QuickSearch:
  Number of matching entries: 0.

You can search over all the following column properties (e.g year, author, title)
Last Update on 03/12/2008
AuthorTitleYearJournal/ProceedingsBibTeX typeDOI/URL/PDF
Mario Frank, Matthias Plaue, H. R. U. K. B. J. & Hamprecht, F. A. Theoretical and Experimental Error Analysis of Continuous-Wave Time-Of-Flight Range Cameras 2009 Optical Engineering   article [PDF]  
BibTeX:
@article{frank09_TOFerror,
  author = {Mario Frank, Matthias Plaue, Holger Rapp, Ullrich Köthe, Bernd Jähne and Fred A. Hamprecht},
  title = {Theoretical and Experimental Error Analysis of Continuous-Wave Time-Of-Flight Range Cameras},
  journal = {Optical Engineering},
  publisher = {Society of Photo-Optical Instrumentation Engineers (SPIE)},
  year = {2009},
  volume = {48},
  number = {1}
}
Brand, M. & Pletscher, P. A conditional random field for automatic photo editing 2008 Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)   article  
BibTeX:
@article{BrandP:CRFAutomaticPhotoEditing:2008,
  author = {Matthew Brand and Patrick Pletscher},
  title = {A conditional random field for automatic photo editing},
  booktitle = {Proceedings of the {IEEE} Conference on Computer Vision and Pattern Recognition ({CVPR})},
  year = {2008}
}
Cerf, M., Harel, J., Einhäuser, W. & Koch, C. Predicting human gaze using low-level saliency combined with face detection 2008 Advances in Neural Information Processing Systems (NIPS) 20   inproceedings URL  
Abstract: Under natural viewing conditions, human observers shift their gaze to allocate

processing resources to subsets of the visual input. Many computational models

try to predict such voluntary eye and attentional shifts. Although the important

role of high level stimulus properties (e.g., semantic information) in search

stands undisputed, most models are based on low-level image properties. We here

demonstrate that a combined model of face detection and low-level saliency significantly

outperforms a low-level model in predicting locations humans fixate on,

based on eye-movement recordings of humans observing photographs of natural

scenes, most of which contained at least one person. Observers, even when not instructed

to look for anything particular, fixate on a face with a probability of over

80% within their first two fixations; furthermore, they exhibit more similar scanpaths

when faces are present. Remarkably, our model’s predictive performance in

images that do not contain faces is not impaired, and is even improved in some

cases by spurious face detector responses.

BibTeX:
@inproceedings{Cerf2008,
  author = {Moran Cerf and Jonathan Harel and Wolfgang Einh\"auser and Christof Koch},
  title = {Predicting human gaze using low-level saliency combined with face detection},
  booktitle = {Advances in Neural Information Processing Systems (NIPS) 20},
  publisher = {MIT Press},
  year = {2008},
  note = {(accepted)},
  url = {http://books.nips.cc/papers/files/nips20/NIPS2007_1074.pdf}
}
Einhäuser, W., Stout, J., Koch, C. & Carter, O. Pupil dilation reflects perceptual selection and predicts subsequent stability in perceptual rivalry 2008 Proceedings of the National Academy of Sciences of the United States of America   article URL  
Abstract: During sustained viewing of an ambiguous stimulus, an individual’s perceptual experience will generally switch between the different possible alternatives rather than stay fixed on one interpretation (perceptual rivalry). Here, we measured pupil diameter while subjects viewed different ambiguous visual and auditory stimuli. For all stimuli tested, pupil diameter increased just before the reported perceptual switch and the relative amount of dilation before this switch was a significant predictor of the subsequent duration of perceptual stability. These results could not be explained by blink or eye-movement effects, the motor response or stimulus driven changes in retinal input. Because pupil dilation reflects levels of norepinephrine (NE) released from the locus coeruleus (LC), we interpret these results as suggestive that the LC–NE complex may play the same role in perceptual selection as in behavioral decision making.
BibTeX:
@article{einhauser.stout.ea:pupil,
  author = {Wolfgang Einh\"auser and James Stout and Christof Koch and Olivia Carter},
  title = {Pupil dilation reflects perceptual selection and predicts subsequent stability in perceptual rivalry},
  journal = {Proceedings of the National Academy of Sciences of the United States of America},
  year = {2008},
  volume = {105},
  number = {5},
  pages = {1704-1709},
  url = {http://www.pnas.org/cgi/doi/10.1073/pnas.0707727105}
}
Einhäuser, W., Rutishauser, U. & Koch, C. Task-demands can immediately reverse the effects of sensory-driven saliency in complex visual stimuli 2008 Journal of Vision   article  
Abstract: In natural vision both stimulus features and task demands affect an observer’s attention. However, the relationship between sensory-driven (“bottom-up”) and task-dependent (“top-down”) factors remains controversial: Can task-demands counteract strong sensory signals fully, quickly and irrespective of bottom-up features? To measure attention under naturalistic conditions, we recorded eye-movements in human observers, while they viewed photographs of outdoor scenes. In the first experiment, smooth modulations of contrast biased the stimuli’s sensory-driven saliency towards one side. In free-viewing, observers’ eye-positions were immediately biased toward the high-contrast, i.e. high saliency, side. However, this sensory-driven bias disappeared entirely when observers searched for a bull’s eye target embedded with equal probability to either side of the stimulus. When the target always occurred in the low-contrast side, observers’ eye-positions were immediately biased towards this low-saliency side, i.e. the sensory-driven bias reversed. Hence task-demands do not only override sensory-driven saliency, but also actively countermand it. In a second experiment, a 5-Hz flicker replaced the contrast-gradient. Whereas the bias was less persistent in free viewing, the overriding and reversal took longer to deploy. Hence, insufficient sensory-driven saliency cannot account for the bias reversal. In a third experiment, subjects searched for a spot of locally increased contrast (“oddity”) instead of the bull’s eye (“template”). In contrast to the other conditions, a slight sensory-driven free-viewing bias prevails in this condition. In a forth experiment we demonstrate that at known locations template targets are detected faster than oddity targets, suggesting that the former induce a stronger top-down drive when used as search targets. Taken together, task demands can override sensory-driven saliency in complex visual stimuli almost immediately, and the extent of overriding depends on the search target and the overridden feature, but not on the latter’s free-viewing saliency.
BibTeX:
@article{Einhauser2008,
  author = {W Einh{\"a}user and U Rutishauser and C Koch},
  title = {Task-demands can immediately reverse the effects of sensory-driven saliency in complex visual stimuli},
  journal = {Journal of Vision},
  year = {2008},
  note = {in press}
}
Fürnstahl, P., Fuchs, T., Schweizer, A., Nagy, L., Székely, G. & Harders, M. Automatic and Robust Forearm Segmentation using Graph Cuts 2008 Biomedical Imaging: From Nano to Macro, 2008. ISBI 2008. 5th IEEE International Symposium on   inproceedings  
BibTeX:
@inproceedings{,
  author = {F\"urnstahl, P. and Fuchs, T. and Schweizer, A. and Nagy, L. and Sz\'ekely, G. and M. Harders},
  title = {Automatic and Robust Forearm Segmentation using Graph Cuts},
  booktitle = {Biomedical Imaging: From Nano to Macro, 2008. ISBI 2008. 5th IEEE International Symposium on},
  publisher = {IEEE},
  year = {2008},
  pages = {77--80}
}
Frank, M., Basin, D. & Buhmann, J. M. A Class of Probabilistic Models for Role Engineering 2008 15th ACM Conference on Computer and Communications Security (CCS 2008)   inproceedings [PDF]  
BibTeX:
@inproceedings{frank.ea:class:2008,
  author = {Mario Frank and David Basin and Joachim M. Buhmann},
  title = {A Class of Probabilistic Models for Role Engineering},
  booktitle = {15th ACM Conference on Computer and Communications Security (CCS 2008)},
  publisher = {ACM},
  year = {2008}
}
Gluz O, Wild P, M. R. D. R. T. E. M. S. S. G. D. E. F. T. H. A. G. A. F. M. P. C. N. U. H. A. Nuclear Karyopherin ?2 expression predicts poor survival in patients with advanced breast cancer irrespective of treatment intensity. 2008 International Journal of Cancer 2008.   article  
BibTeX:
@article{,
  author = {Gluz O, Wild P, Meiler R, Diallo-Danebrock R, Ting E, Mohrmann S, Schuett G, Dahl E, Fuchs T, Herr A, Gaumann A, Frick M, Poremba C,Nitz UA, Hartmann A.},
  title = {Nuclear Karyopherin ?2 expression predicts poor survival in patients with advanced breast cancer irrespective of treatment intensity.},
  journal = { International Journal of Cancer 2008.},
  year = {2008},
  number = {IF 4,693}
}
Moh, Y. & Buhmann, J. Kernel Expansion for Online Preference Tracking 2008 9th International Conference on Music Information Retrieval   inproceedings  
BibTeX:
@inproceedings{Moh2008b,
  author = {Yvonne Moh and Joachim Buhmann},
  title = {Kernel Expansion for Online Preference Tracking},
  booktitle = {9th International Conference on Music Information Retrieval},
  year = {2008}
}
Moh, Y., Einhäuser, W. & Buhmann, J. Automatic Detection of Learnability under Unreliable and Sparse User Feedback 2008 Pattern Recognition--DAGM 2008   inproceedings [PDF]  
BibTeX:
@inproceedings{Moh2008a,
  author = {Yvonne Moh and Wolfgang Einh\"auser and Joachim Buhmann},
  title = {Automatic Detection of Learnability under Unreliable and Sparse User Feedback},
  booktitle = {Pattern Recognition--DAGM 2008},
  publisher = {Springer},
  year = {2008}
}
Moh, Y., Orbanz, P. & Buhmann, J. M. Music Preference Learning with Partial Information 2008 Proccedings of International Conference on Acoustic, Speech, and Signal Processing   inproceedings  
BibTeX:
@inproceedings{Moh2008,
  author = {Y. Moh and P. Orbanz and J. M. Buhmann},
  title = {Music Preference Learning with Partial Information},
  booktitle = {Proccedings of International Conference on Acoustic, Speech, and Signal Processing},
  year = {2008}
}
Niesen, U. H., Pellet, J. P. & Elisseeff, A. Explanation Trees for Causal Bayesian Networks 2008 Proceedings of the 24th Conference in Uncertainty in Artificial Intelligence   inproceedings [PDF]  
BibTeX:
@inproceedings{nielsen08uai,
  author = {Ulf Holm Niesen and Jean Philippe Pellet and Andr¥'{e} Elisseeff},
  title = {Explanation Trees for Causal {B}ayesian Networks},
  booktitle = {Proceedings of the 24th Conference in Uncertainty in Artificial Intelligence},
  publisher = {AUAI Press},
  year = {2008},
  pages = {427--434}
}
Orbanz, P. & Buhmann, J. M. Nonparametric Bayesian Image Segmentation 2008 International Journal of Computer Vision   article  
BibTeX:
@article{orbanz.buhmann:nonparametric,
  author = {P. Orbanz and J.~M. Buhmann},
  title = {Nonparametric {B}ayesian Image Segmentation},
  journal = {International Journal of Computer Vision},
  year = {2008},
  volume = {77}
}
Pellet, J. P. & Elisseeff, A. Using Markov blankets for Causal Structure Learning 2008 Journal of Machine Learning Research   article [PDF]  
Abstract: We show how a generic feature-selection algorithm returning strongly relevant variables can be turned into a causal structure-learning algorithm. We prove this under the Faithfulness assumption for the data distribution. In a causal graph, the strongly relevant variables for a node X are its parents, children, and children’s parents (or spouses), also known as the Markov blanket of X. Identifying the spouses leads to the detection of the V-structure patterns and thus to causal orientations. Repeating the task for all variables yields a valid partially oriented causal graph. We ?rst show an ef?cient way to identify the spouse links. We then perform several experiments in the continuous domain using the Recursive Feature Elimination feature-selection algorithm with Support Vector Regression and empirically verify the intuition of this direct (but computationally expensive) approach. Within the same framework, we then devise a fast and consistent algorithm, Total Conditioning (TC), and a variant, TCbw, with an explicit backward feature-selection heuristics, for Gaussian data. After running a series of comparative experiments on ?ve arti?cial networks, we argue that Markov blanket algorithms such as TC/TCbw or Grow-Shrink scale better than the reference PC algorithm and provides higher structural accuracy.
BibTeX:
@article{pellet08jmlr,
  author = {Jean Philippe Pellet and Andr\'{e} Elisseeff},
  title = {Using {M}arkov blankets for Causal Structure Learning},
  journal = {Journal of Machine Learning Research},
  year = {2008},
  volume = {9},
  pages = {1295--1342}
}
Sigg, C. D. & Buhmann, J. M. Expectation-Maximization for Sparse and Non-Negative PCA 2008 Proc. 25th International Conference on Machine Learning,   inproceedings [PDF]  
BibTeX:
@inproceedings{chrsigg2008a,
  author = {Christian D. Sigg and Joachim M. Buhmann},
  title = {Expectation-Maximization for Sparse and Non-Negative PCA},
  booktitle = {Proc. 25th International Conference on Machine Learning,},
  year = {2008}
}
Stefan Steiniger, Tilman Lange, D. B. R. W. An Approach for the Classification of Urban Building Structures Based on Discriminant Analysis Techniques 2008 Transactions in GIS   article  
BibTeX:
@article{,
  author = {Stefan Steiniger, Tilman Lange, Dirk Burghardt, Robert Weibel},
  title = {An Approach for the Classification of Urban Building Structures Based on Discriminant Analysis Techniques},
  journal = {Transactions in GIS},
  year = {2008},
  volume = {12(1)},
  pages = {31-59}
}
Streich, A. P. & Buhmann, J. M. Classification of Multi-Labeled Data: A Generative Approach 2008 ECML 2008   inproceedings  
Abstract: Multi-label classification assigns a data item to one or several classes. This problem of multiple labels arises in fields like acoustic and visual scene analysis, news reports and medical diagnosis. In a generative framework, data with multiple labels can be interpreted as additive mixtures of emissions of the individual sources. We propose a deconvolution approach to estimate the individual contributions of each source to a given data item. Similarly, the distributions of multi-label data are computed based on the source distributions. In experiments with synthetic data, the novel approach is compared to existing models and yields more accurate parameter estimates, higher classification accuracy and ameliorated generalization to previously unseen label sets. These improvements are most pronounced on small training data sets. Also on real world acoustic data, the algorithm outperforms other generative models, in particular on small training data sets.
BibTeX:
@inproceedings{streich.buhmann:multilabel_generative,
  author = {Andreas P. Streich and Joachim M. Buhmann},
  title = {Classification of Multi-Labeled Data: A Generative Approach},
  booktitle = {ECML 2008},
  year = {2008},
  note = {(to be published)}
}
Thomas Fuchs, Peter Wild, H. M. J. B. Computational Pathology Analysis of Tissue Microarrays Predicts Survival of Renal Clear Cell Carcinoma Patients 2008 Medical Image Computing and Computer-Assisted Intervention - MICCAI 2008   inproceedings  
BibTeX:
@inproceedings{,
  author = {Thomas Fuchs, Peter Wild, Holger Moch, Joachim Buhmann},
  title = {Computational Pathology Analysis of Tissue Microarrays Predicts Survival of Renal Clear Cell Carcinoma Patients},
  booktitle = {Medical Image Computing and Computer-Assisted Intervention - MICCAI 2008},
  publisher = {Springer},
  year = {2008}
}
Thomas Fuchs, Tilman Lange, P. W. H. M. J. B. Weakly Supervised Cell Nuclei Detection and Segmentation on Tissue Microarrays of Renal Cell Carcinoma 2008 Pattern Recognition   inproceedings DOI  
BibTeX:
@inproceedings{,
  author = {Thomas Fuchs, Tilman Lange, Peter Wild, Holger Moch, Joachim Buhmann},
  title = {Weakly Supervised Cell Nuclei Detection and Segmentation on Tissue Microarrays of Renal Cell Carcinoma},
  booktitle = {Pattern Recognition},
  publisher = {Springer Berlin / Heidelberg},
  year = {2008},
  volume = {5096/2008},
  pages = {173-182},
  doi = {10.1007/978-3-540-69321-}
}
Busse, L., Orbanz, P. & Buhmann, J. M. Cluster Analysis of Heterogeneous Rank Data 2007 ICML 2007   inproceedings [PDF]  
Abstract: Cluster analysis of ranking data, which occurs in consumer questionnaires, voting forms or other inquiries of preferences, attempts to identify typical groups of rankchoices. Empirically measured rankings are often incomplete, i.e. different numbers of filled rank positions cause heterogeneity in the data. We propose a mixture approach forclustering of heterogeneous rank data. Rankings of different lengths can be described and compared by means of a single probabilistic model. A maximum entropy approach avoidshidden assumptions about missing rank positions. Parameter estimators and an efficient EM algorithm for unsupervised inference are derived for the ranking mixture model. Experiments on both synthetic data and real-world data demonstrate significantly improved parameter estimates on heterogeneous data when the incomplete rankings areincluded in the inference process.
BibTeX:
@inproceedings{Busse2007a,
  author = {Ludwig Busse and Peter Orbanz and Joachim M. Buhmann},
  title = {Cluster Analysis of Heterogeneous Rank Data},
  booktitle = {ICML 2007},
  year = {2007},
  note = {(in press)}
}
Cerf, M., Cleary, D. R., Peters, R. J., Einhäuser, W. & Koch, C. Observers are Consistent when Rating Image Conspicuity 2007 Vision Research   article URL [PDF]  
Abstract: Human perception of an image’s conspicuity depends on the stimulus itself and the observer’s semantic interpretation. We investigated the relative contribution of the former, sensory-driven, component. Participants viewed sequences of images from five different classes—fractals, overhead satellite imagery, grayscale and colored natural scenes, and magazine covers—and graded each numerically according to its perceived conspicuity. We found significant consistency in this rating within and between observers for all image categories. In a subsequent recognition memory test, performance was significantly above chance for all categories, with the weakest memory for satellite imagery, and reaching near ceiling for magazine covers. When repeating the experiment after one year, ratings remained consistent within each observer and category, despite the absence of explicit scene memory. Our findings suggest that the rating of image conspicuity is driven by image-immanent, sensory factors common to all observers.
BibTeX:
@article{Cerf2007,
  author = {Moran Cerf and Daniel R. Cleary and Robert J. Peters and Wolfgang Einh\"auser and Christof Koch},
  title = {Observers are Consistent when Rating Image Conspicuity},
  journal = {Vision Research},
  year = {2007},
  volume = {47},
  number = {24},
  pages = {3052--3060},
  url = {http://dx.doi.org/10.1016/j.visres.2007.06.025}
}
Einhäuser, W., Koch, C. & Makeig, S. The duration of the attentional blink in natural scenes depends on stimulus category 2007 Vision Research   article URL [PDF]  
Abstract: Humans comprehend the ''gist'' of even a complex natural scene within a small fraction of a second. If, however, observers are asked to detect targets in a sequence of rapidly presented items, recognition of a target succeeding another target by about a third of a second is severely impaired, the ''attentional blink'' (AB) [Raymond, J. E., Shapiro, K. L., & Arnell, K. M. (1992). Temporary suppression of visual processing in an RSVP task: an attentional blink? Journal of Experimental Psychology. Human Perception and Performance, 18, 849-860].Since most experiments on the AB use well controlled but artificial stimuli, the question arises whether the same phenomenon occurs for complex, natural stimuli, and if so, whether its specifics depend on stimulus category. Here we presented rapid sequences of complex stimuli (photographs of objects, scenes and faces) and asked observers to detect and remember items of a specific category (either faces, watches, or both). We found a consistent AB for both target categories but the duration of the AB depended on the target category.
BibTeX:
@article{Einhauser2007a,
  author = {Wolfgang Einh\"auser and Christof Koch and Scott Makeig},
  title = {The duration of the attentional blink in natural scenes depends on stimulus category},
  journal = {Vision Research},
  year = {2007},
  volume = {47},
  number = {5},
  pages = {597-607},
  url = {http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6T0W-4MYF60S-6&_user=791130&_coverDate=03/31/2007&_rdoc=1&_fmt=&_orig=search&_sort=d&view=c&_acct=C000043379&_version=1&_urlVersion=0&_userid=791130&md5=a7a9c435d306221ebe934dc213738dbf}
}
Einhäuser, W., Mundhenk, T. N., Baldi, P., Koch, C. & Itti, L. A Bottom-Up Model of Spatial Attention Predicts Human Error Patterns in Rapid Scene Recognition 2007 Journal of Vision   article URL [PDF]  
Abstract: Humans demonstrate a peculiar ability to detect complex targets in rapidly presented natural scenes. Recent studies suggest that (nearly) no focal attention is required for overall performance in such tasks. Little is known, however, of how detection performance varies from trial to trial and which stages in the processing hierarchy limit performance: bottom–up visual processing (attentional selection and/or recognition) or top–down factors (e.g., decision-making, memory, or alertness fluctuations)? To investigate the relative contribution of these factors, eight human observers performed an animal detection task in natural scenes presented at 20 Hz. Trial-by-trial performance was highly consistent across observers, far exceeding the prediction of independent errors. This consistency demonstrates that performance is not primarily limited by idiosyncratic factors but by visual processing. Two statistical stimulus properties, contrast variation in the target image and the information-theoretical measure of “surprise” in adjacent images, predict performance on a trial-by-trial basis. These measures are tightly related to spatial attention, demonstrating that spatial attention and rapid target detection share common mechanisms. To isolate the causal contribution of the surprise measure, eight additional observers performed the animal detection task in sequences that were reordered versions of those all subjects had correctly recognized in the first experiment. Reordering increased surprise before and/or after the target while keeping the target and distractors themselves unchanged. Surprise enhancement impaired target detection in all observers. Consequently, and contrary to several previously published findings, our results demonstrate that attentional limitations, rather than target recognition alone, affect the detection of targets in rapidly presented visual sequences.
BibTeX:
@article{Einhauser2007b,
  author = {Wolfgang Einh\"auser and T. Nathan Mundhenk and Pierre Baldi and Christof Koch and Laurent Itti},
  title = {A Bottom-Up Model of Spatial Attention Predicts Human Error Patterns in Rapid Scene Recognition},
  journal = {Journal of Vision},
  year = {2007},
  volume = {7},
  number = {10},
  pages = {1-13},
  url = {http://www.journalofvision.org/7/10/6/}
}
Einhäuser, W., Schumann, F., Bardins, S., Bartl, K., Böning, G., Schneider, E. & König., P. Human eye-head co-ordination in natural exploration 2007 Network: Computation in Neural Systems   article URL [PDF]  
Abstract: During natural behavior humans continuously adjust their gaze by moving head and eyes,yielding rich dynamics of the retinal input. Sensory coding models, however, typically assumevisual input as smooth or a sequence of static images interleaved by volitional gaze shifts.Are these assumptions valid during free exploration behavior in natural environments? Weused an innovative technique to simultaneously record gaze and head movements in humans,who freely explored various environments (forest, train station, apartment). Most movementsoccur along the cardinal axes, and the predominance of vertical or horizontal movementsdepends on the environment. Eye and head movements co-occur more frequently than theirindividual statistics predicts under an independence assumption. The majority of cooccurringmovements point in opposite directions, consistent with a gaze-stabilizing role ofeye movements. Nevertheless, a substantial fraction of eye movements point in the samedirection as co-occurring head movements. Even under the very most conservativeassumptions, saccadic eye movements alone cannot account for these synergistic movements.Hence nonsaccadic eye movements that interact synergistically with head movements toadjust gaze cannot be neglected in natural visual input. Natural retinal input is continuouslydynamic, and cannot be faithfully modeled as a mere sequence of static frames withinterleaved large saccades.
BibTeX:
@article{Einhauser2007,
  author = {Wolfgang Einh\"auser and Frank Schumann and Stanislavs Bardins and Klaus Bartl and Guido B\"oning and Erich Schneider and Peter K\"onig.},
  title = {Human eye-head co-ordination in natural exploration},
  journal = {Network: Computation in Neural Systems},
  year = {2007},
  volume = {18},
  number = {3},
  pages = {267-297},
  url = {http://www.informaworld.com/openurl?genre=article&issn=0954-898X&volume=18&issue=3&spage=267}
}
Fischer, B., Roth, V. & Buhmann, J. Time-Series Alignment by Non-Negative Multiple Generalized Canonical Correlation Analysis 2007 BMC Bioinformatics   article URL  
Abstract: bf Background: Quantitative analysis of differential protein expressions requires to align temporal elution measurements from liquid chromatography coupled to mass spectrometry (LC/MS). We propose multiple Canonical Correlation Analysis (mCCA) as a method to align the non-linearly distorted time scales of repeated LC/MS experiments in a robust way.\ bf Results: Multiple canonical correlation analysis is able to map several time series to a consensus time scale. The alignment function is learned in a supervised fashion. We compare our approach with previously published methods for aligning mass spectrometry data on a large proteomics dataset. The proposed method significantly increases the number of proteins that are identified as being differentially expressed in different biological samples.\ bf Conclusion: Jointly aligning multiple liquid chromatography/mass spectrometry samples by mCCA substantially increases the detection rate of potential bio-markers which significantly improves the interpretability of LC/MS data.
BibTeX:
@article{fischer.roth.ea:time-series,
  author = {Bernd Fischer and Volker Roth and Joachim Buhmann},
  title = {Time-Series Alignment by Non-Negative Multiple Generalized Canonical Correlation Analysis},
  journal = {BMC Bioinformatics},
  year = {2007},
  volume = {8},
  number = {S10},
  pages = {S4},
  url = {http://dx.doi.org/10.1186/1471-2105-8-S10-S4}
}
Fischer, B., Roth, V. & Buhmann, J. M. Time-Series Alignment by Non-Negative Multiple Generalized Canonical Correlation Analysis 2007 Computational Intelligence Methods for Bioinformatics and Biostatistics   inproceedings URL  
Abstract: For a quantitative analysis of differential protein expression, one has to overcome the problem of aligning time series of measurements from liquid chromatography coupled to mass spectrometry. When repeating experiments one typically observes that the time axis is deformed in a non-linear way. In this paper we propose a technique to align the time series based on generalized canonical correlation analysis (GCCA) for multiple datasets. The monotonicity constraint in time series alignment is incorporated in the GCCA algorithm. The alignment function is learned both in a supervised and a semi-supervised fashion. We compare our approach with previously published methods for aligning mass spectrometry data on a large proteomics dataset.
BibTeX:
@inproceedings{fischerRothBuhmann:07a,
  author = {Bernd Fischer and Volker Roth and Joachim M. Buhmann},
  title = {Time-Series Alignment by Non-Negative Multiple Generalized Canonical Correlation Analysis},
  booktitle = {Computational Intelligence Methods for Bioinformatics and Biostatistics},
  publisher = {Springer},
  year = {2007},
  volume = {4578},
  pages = {505--511},
  url = {http://dx.doi.org/10.1007/978-3-540-73400-0_64}
}
Frey, H., König, P. & Einhäuser, W. The Role of First- and Second Order Stimulus Features for Human Overt Attention 2007 Perception and Psychophysics   article [PDF]  
Abstract: When processing complex visual input, human observers sequentially allocate their attention to different subsets of the stimulus. What are the mechanisms and strategiesthat guide this selection process? Here we investigate the influence of various stimulus features on human overt attention, i.e. attention related to shifts of gaze, innatural color images and modified versions thereof. The modifications, systematic changes of hue across the entire image, influence only the global appearance of thestimuli, while leaving the local features under investigation unaffected. We demonstrate that the modifications reduce the subjective interpretation of a stimulus as ''natural'' consistently across observers. By analyzing fixations, we find that firstorder features, such as luminance-contrast, saturation and color-contrast along either of the cardinal axes, are correlated to overt attention in the modified images. In contrast, no such correlation is found in unmodified outdoor images. Second-order luminance-contrast (''texture-contrast'') is correlated to overt attention in all conditions. However, none of the second-order color-contrasts are correlated to overt attention in unmodified images while one of the second-order color-contrasts exhibits a significant correlation in modified images. These findings, on the one hand, imply that higher order bottom-up effects, namely those of second-order luminance-contrast, may partly account for human overt attention. On the other hand, these results also demonstrate that global images properties, which correlate to the subjective impression of a scene as being ''natural'', affect the guidance of human overt attention.
BibTeX:
@article{Frey2007,
  author = {Hans-Peter Frey and Peter K\"onig and Wolfgang Einh\"auser},
  title = {The Role of First- and Second Order Stimulus Features for Human Overt Attention},
  journal = {Perception and Psychophysics},
  year = {2007},
  volume = {69},
  number = {2},
  pages = {153--161}
}
Grossmann$^, J., Fischer$^, B., Bärenfaller, K., Owitil, J., Buhmann, J. M., Gruissem, W. & Baginsky, S. A workflow to increase the detection rate of proteins from un-sequenced organisms inhigh-throughput proteomics experiments 2007 Proteomics   article URL  
Abstract: We present and evaluate a strategy for the mass spectrometric identification of proteins from organisms for which no genome sequence information is available that incorporates cross-species information from sequenced organisms. The presented method combines spectrum quality scoring, de novo sequencing and error tolerant BLAST searches and is designed to decrease input data complexity. Spectral quality scoring reduces the number of investigated mass spectra without a loss of information. Stringent quality-based selection and the combination of different de novo sequencing methods substantially increase the catalog of significant peptide alignments. The de novo sequences passing a reliability filter are subsequently submitted to error tolerant BLAST searches and MS-BLAST hits are validated by a sampling technique. With the described workflow, we identified up to 20% more groups of homologous proteins in proteome analyses with organisms whose genome is not sequenced than by state-of-the-art database searches in an Arabidopsis thaliana database. We consider the novel data analysis workflow an excellent screening method to identify those proteins that evade detection in proteomics experiments as a result of database constraints.
BibTeX:
@article{grossmann.fischer.ea:workflow,
  author = {Jonas Grossmann$^\ast$ and Bernd Fischer$^\ast$ and Katja B\"arenfaller and Judith Owitil and Joachim M. Buhmann and Wilhelm Gruissem and Sacha Baginsky},
  title = {A workflow to increase the detection rate of proteins from un-sequenced organisms inhigh-throughput proteomics experiments},
  journal = {Proteomics},
  year = {2007},
  volume = {7},
  number = {23},
  pages = {4245 -- 4254},
  note = {$^\ast$authors contributed equally},
  url = {http://dx.doi.org/10.1002/pmic.200700474}
}
Kaynig, V., Fischer, B., Wepf, R. & Buhmann, J. M. Fully Automatic Registration of Electron Microscopy Images with High and Low Resolution 2007 Microscopy and Microanalysis   inproceedings URL  
BibTeX:
@inproceedings{Kaynig2007,
  author = {Verena Kaynig and Bernd Fischer and Roger Wepf and Joachim M. Buhmann},
  title = {Fully Automatic Registration of Electron Microscopy Images with High and Low Resolution},
  booktitle = {Microscopy and Microanalysis},
  year = {2007},
  url = {http://journals.cambridge.org/action/displayAbstract?fromPage=online&aid=1213224#}
}
Lange, T. & Buhmann, J. M. Regularized Data Fusion Improves Image Segmentation 2007 Pattern Recognition - Symposium of the DAGM 2007   inproceedings  
BibTeX:
@inproceedings{Lange2007,
  author = {Tilman Lange and Joachim M. Buhmann},
  title = {Regularized Data Fusion Improves Image Segmentation},
  booktitle = {Pattern Recognition - Symposium of the DAGM 2007},
  publisher = {Springer},
  year = {2007},
  volume = {4713},
  pages = {234-243}
}
Lange, T. & Buhmann, J. M. Kernel-Based Grouping of Histogramm Data 2007 ECML 2007   inproceedings  
BibTeX:
@inproceedings{Lange2007b,
  author = {Tilman Lange and Joachim M. Buhmann},
  title = {Kernel-Based Grouping of Histogramm Data},
  booktitle = {ECML 2007},
  year = {2007},
  note = {(to be published)}
}
Ommer, B. & Buhmann, J. M. Learning the Compositional Nature of Visual Objects 2007 CVPR 2007   inproceedings  
BibTeX:
@inproceedings{Ommer2007,
  author = {Bj\"orn Ommer and Joachim M. Buhmann},
  title = {Learning the Compositional Nature of Visual Objects},
  booktitle = {CVPR 2007},
  year = {2007},
  note = {(in press)}
}
Ommer, B. & Buhmann, J. M. Compositional Object Recognition, Segmentation, and Tracking in Video 2007 International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR'07)   inproceedings [PDF]  
BibTeX:
@inproceedings{Ommer2007a,
  author = {Bj\"orn Ommer and Joachim M. Buhmann},
  title = {Compositional Object Recognition, Segmentation, and Tracking in Video},
  booktitle = {International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR'07)},
  publisher = {Springer},
  year = {2007},
  volume = {4679},
  pages = {318-333}
}
Orbanz, P., Braendle, S. & Buhmann, J. M. Bayesian Order-Adaptive Clustering for Video Segmentation 2007 EMMCVPR 2007   inproceedings [PDF]  
Abstract: Video segmentation requires the partitioning of a series ofimages into groups that are both spatially coherent and smooth along the time axis. We formulate segmentation as a Bayesian clustering problem. Context information is propagated over time by a conjugate structure. The level of segment resolution is controlled by a Dirichlet process prior. Our contributions include a conjugate nonparametric Bayesian model for clustering in multivariate time series, a MCMC inference algorithm, and a multiscale sampling approach for Dirichlet process mixture models. The multiscale algorithm is applicable to data with a spatial structure.The method is tested on synthetic data and on videos from the MPEG4 benchmark set.
BibTeX:
@inproceedings{Orbanz2007b,
  author = {Peter Orbanz and Samuel Braendle and Joachim M. Buhmann},
  title = {Bayesian Order-Adaptive Clustering for Video Segmentation},
  booktitle = {EMMCVPR 2007},
  publisher = {Springer},
  year = {2007},
  pages = {334--349}
}
Pellet, J. P. & Elisseeff, A. A Partial Correlation-Based Algorithm for Causal Structure Discovery with Continuous Variables 2007 Advances in Intelligent Data Analysis VII, 7th International Symposium on Intelligent Data Analysis   inproceedings [PDF]  
Abstract: We present an algorithm for causal structure discovery suited in the presence of continuous variables. We test a version based on partial correlation that is able to recover the structure of a recursive linear equations model and compare it to the well-known PC algorithm on large networks. PC is generally outperformed in run time and number of structural errors.
BibTeX:
@inproceedings{pellet07ida,
  author = {Jean Philippe Pellet and Andr\'{e} Elisseeff},
  title = {A Partial Correlation-Based Algorithm for Causal Structure Discovery with Continuous Variables},
  booktitle = {Advances in Intelligent Data Analysis VII, 7th International Symposium on Intelligent Data Analysis},
  year = {2007},
  pages = {229--239}
}
Roos$^, F., Jacob$^, R., Grossmann, J., Fischer, B., Buhmann, J., Gruissem, W., Baginsky, S. & Widmayer, P. PepSplice: Cache-Efficient Search Algorithms for Comprehensive Identification of Tandem Mass Spectra 2007 Bioinformatics   article URL  
Abstract: bf Motivation: Tandem mass spectrometry allows for high-throughput identification of complex protein samples. Searching tandem mass spectra against sequence databases is the main analysis method nowadays. Since many peptide variations are possible, including them in the search space seems only logical. However, the search space usually grows exponentially with the number of independent variations and may therefore overwhelm computational resources.\ bf Results: We provide fast, cache-efficient search algorithms to screen large peptide search spaces including non-tryptic peptides, whole genomes, dozens of posttranslational modifications, unannotated point mutations and even unannotated splice sites. All these search spaces can be screened simultaneously. By optimizing the cache usage, we achieve a calculation speed that closely approaches the limits of the hardware. At the same time, we control the size of the overall search space by limiting the combinations of variations that can co-occur on the same peptide. Using a hypergeometric scoring scheme, we applied these algorithms to a dataset of 1 420 632 spectra. We were able to identify a considerable number of peptide variations within a modest amount of computing time on standard desktop computers.\ bf Availability: PepSplice is available as a C++ application for Linux, Windows and OSX at www.ti.inf.ethz.ch/pw/software/pepsplice/. It is open source under the revised BSD license.
BibTeX:
@article{roos.jacob.ea:pepsplice,
  author = {Franz Roos$^\ast$ and Riko Jacob$^\ast$ and Jonas Grossmann and Bernd Fischer and Joachim Buhmann and Wilhelm Gruissem and Sacha Baginsky and Peter Widmayer},
  title = {PepSplice: Cache-Efficient Search Algorithms for Comprehensive Identification of Tandem Mass Spectra},
  journal = {Bioinformatics},
  year = {2007},
  volume = {23},
  number = {22},
  pages = {3016--3023},
  note = {$^\ast$authors contributed equally},
  url = {http://dx.doi.org/10.1093/bioinformatics/btm417}
}
Roth, V. & Fischer, B. The kernelHMM: Learning Kernel Combinations in Structured Output Domains 2007 Pattern Recognition (29th DAGM Symposium)   inproceedings URL  
Abstract: We present a model for learning convex kernel combinations in classification problems with structured output domains. The main ingredient is a hidden Markov model which forms a layered directed graph. Each individual layer represents a multilabel version of nonlinear kernel discriminant analysis for estimating the emission probabilities. These kernel learning machines are equipped with a mechanism for finding convex combinations of kernel matrices. The resulting kernelHMM can handle multiple partial paths through the label hierarchy in a consistent way. Efficient approximation algorithms allow us to train the model to large-scale learning problems. Applied to the problem of document categorization, the method exhibits excellent predictive performance.
BibTeX:
@inproceedings{Roth2007,
  author = {Volker Roth and Bernd Fischer},
  title = {The kernelHMM: Learning Kernel Combinations in Structured Output Domains},
  booktitle = {Pattern Recognition (29th DAGM Symposium)},
  publisher = {Springer},
  year = {2007},
  volume = {4713},
  pages = {436--445},
  url = {http://www.springerlink.com/content/r015664w85831633/fulltext.pdf}
}
Roth, V. & Fischer, B. Improved Functional Prediction of Proteins by Learning Kernel Combinations in Multilabel Settings 2007 BMC Bioinformatics   article URL  
Abstract: bf Background: We develop a probabilistic model for combining kernel matrices to predict the function of proteins. It extends previous approaches in that it can handle multiple labels which naturally appear in the context of protein function.\ bf Results: Explicit modeling of multilabels significantly improves the capability of learning protein function from multiple kernels. The performance and the interpretability of the inference model are further improved by simultaneously predicting the subcellular localization of proteins and by combining pairwise classifiers to consistent class membership estimates.\ bf Conclusion: For the purpose of functional prediction of proteins, multilabels provide valuable information that should be included adequately in the training process of classifiers. Learning of functional categories gains from co-prediction of subcellular localization. Pairwise separation rules allow very detailed insights into the relevance of different measurements like sequence, structure, interaction data, or expression data. A preliminary version of the software can be downloaded from http://www.inf.ethz.ch/personal/vroth/KernelHMM/
BibTeX:
@article{rothFischer:07a,
  author = {Volker Roth and Bernd Fischer},
  title = {Improved Functional Prediction of Proteins by Learning Kernel Combinations in Multilabel Settings},
  journal = {BMC Bioinformatics},
  year = {2007},
  volume = {8},
  number = {S2},
  pages = {S12},
  url = {http://dx.doi.org/10.1186/1471-2105-8-S2-S12}
}
Sigg, C., Fischer, B., Ommer, B., Roth, V. & Buhmann, J. Nonnegative CCA for Audiovisual Source Separation 2007 IEEE Workshop on Machine Learning for Signal Processing   inproceedings [PDF]  
Abstract: We present a method for finding correlated components in audio and video signals. The new technique is applied to the task of identifying sources in video and separating them in audio. The concept of canonical correlation analysis is reformulated such that it incorporates nonnegativity and sparsity constraints on the coefficients of projection directions. Nonnegativity ensures that projections are compatible with an interpretation as energy signals. Sparsity ensures that coefficient weight concentrates on individual sources. By finding multiple conjugate directions we finally obtain a component based decomposition of both data modalities. Experiments effectively demonstrate the potential and benefits of this approach.
BibTeX:
@inproceedings{Sigg2007,
  author = {Christian Sigg and Bernd Fischer and Bjoern Ommer and Volker Roth and Joachim Buhmann},
  title = {Nonnegative CCA for Audiovisual Source Separation},
  booktitle = {IEEE Workshop on Machine Learning for Signal Processing},
  publisher = {IEEE Press},
  year = {2007}
}
Zöller, T. & Buhmann, J. M. Robust Image Segmentation using Resampling and Shape Constraints 2007 IEEE Transactions on Pattern Analysis and Machine Intelligence   article  
BibTeX:
@article{Zoller2007,
  author = {Thomas Z\"oller and Joachim M. Buhmann},
  title = {Robust Image Segmentation using Resampling and Shape Constraints},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2007},
  note = {(in press)}
}
Fischer, B., Grossmann, J., Roth, V., Gruissem, W., Baginsky, S. & Buhmann, J. M. Semi-Supervised LC/MS Alignment for Differential Proteomics 2006 Bioinformatics   article URL [PDF]  
Abstract: Motivation: Mass spectrometry (MS) combined with high-performance liquid chromatography (LC) has received considerable attention for high-throughput analysis of proteomes. Isotopic labeling techniques such as ICAT have been successfully applied to derive differential quantitative information for two protein samples, however at the price of significantly increased complexity of the experimental setup. To overcome these limitations, we consider a label-free setting where correspondences between elements of two samples have to be established prior to the comparative analysis. The alignment between samples is achieved by nonlinear robust ridge regression. The correspondence estimates are guided in a semi-supervised fashion by prior information which is derived from sequenced tandem mass spectra.Results: The semi-supervised method for finding correspondences was successfully applied to aligning highly complex protein samples, even if they exhibit large variations due to different biological conditions. A large-scale experiment clearly demonstrates that the proposed method bridges the gap between statistical data analysis and label-free quantitative differential proteomics.Availability: The software will be available on the website http://people.inf.ethz.ch/befische/proteomics
BibTeX:
@article{fischer.grossmann.ea:semi-supervised,
  author = {Bernd Fischer and Jonas Grossmann and Volker Roth and Wilhelm Gruissem and Sacha Baginsky and Joachim M. Buhmann},
  title = {Semi-Supervised LC/MS Alignment for Differential Proteomics},
  journal = {Bioinformatics},
  year = {2006},
  volume = {22},
  number = {14},
  pages = {e132--e140},
  url = {http://dx.doi.org/10.1093/bioinformatics/btl219}
}
Keuchel, J. Multiclass Image Labeling with Semidefinite Programming 2006 Computer Vision -- ECCV 2006   inproceedings  
BibTeX:
@inproceedings{Keuchel2006a,
  author = {Jens Keuchel},
  title = {Multiclass Image Labeling with Semidefinite Programming},
  booktitle = {Computer Vision -- ECCV 2006},
  publisher = {Springer},
  year = {2006},
  volume = {3952},
  pages = {454-467}
}
Keuchel, J. & Küttel, D. Efficient Combination of Probabilistic Sampling Approximations for Robust Image Segmentation 2006 Pattern Recognition (28th DAGM Symposium)   inproceedings  
BibTeX:
@inproceedings{Keuchel2006,
  author = {J. Keuchel and D. K{\"u}ttel},
  title = {Efficient Combination of Probabilistic Sampling Approximations for Robust Image Segmentation},
  booktitle = {Pattern Recognition (28th DAGM Symposium)},
  publisher = {Springer},
  year = {2006},
  volume = {4174},
  pages = {41-50}
}
Lange, T. & Buhmann, J. Fusion of Similarity Data in Clustering 2006 Advances in Neural Information Processing Systems (NIPS) 19   inproceedings  
BibTeX:
@inproceedings{Lange2006,
  author = {Tilman Lange and Joachim Buhmann},
  title = {Fusion of Similarity Data in Clustering},
  booktitle = {Advances in Neural Information Processing Systems (NIPS) 19},
  publisher = {MIT Press},
  year = {2006},
  pages = {723-730}
}
Laub, J., Roth, V., Buhmann, J. M. & Müller, K. On the Information and Representation of Non-Euclidean Pairwise Data 2006 Pattern Recognition   article  
BibTeX:
@article{Laub2006,
  author = {Julian Laub and Volker Roth and Joachim M. Buhmann and Klaus-Robert M\"uller},
  title = {On the Information and Representation of Non-Euclidean Pairwise Data},
  journal = {Pattern Recognition},
  year = {2006},
  volume = {39},
  pages = {1815-1826}
}
Ommer, B. & Buhmann, J. M. Learning Compositional Categorization Models 2006 ECCV'06   inproceedings [PDF]  
Abstract: This contribution proposes a compositional approach to visual object categorization of scenes. Compositions are learned from the Caltech 101 database and form intermediate abstractions of images that are semantically situated between low-level representations and the high-level categorization. Salient regions, which are described by localized feature histograms, are detected as image parts. Subsequently compositions are formed as bags of parts with a locality constraint. After performing a spatial binding of compositions by means of a shape model, coupled probabilistic kernel classifiers are applied thereupon to establish the final image categorization. In contrast to the discriminative training of the categorizer, intermediate compositions are learned in a generative manner yielding relevant part agglomerations, i.e. groupings which are frequently appearing in the dataset while simultaneously supporting the discrimination between sets of categories. Consequently, compositionality simplifies the learning of a complex categorization model for complete scenes by splitting it up into simpler, sharable compositions. The architecture is evaluated on the highly challenging Caltech 101 database which exhibits large intra-category variations. Our compositional approach shows competitive retrieval rates in the range of 53.6±0.88% or, with a multi-scale feature set, rates of 57.8 ± 0.79%.
BibTeX:
@inproceedings{Ommer2006a,
  author = {Bj\"orn Ommer and Joachim M. Buhmann},
  title = {Learning Compositional Categorization Models},
  booktitle = {ECCV'06},
  publisher = {LNCS 3953, Springer},
  year = {2006}
}
Ommer, B., Sauter, M. & Buhmann, J. M. Learning Top-Down Grouping of Compositional Hierarchies for Recognition 2006 CVPR'06 (POCV)   inproceedings [PDF]  
Abstract: The complexity of real world image categorization and scene analysis requires compositional strategies for object representation. This contribution establishes a ompositionalhierarchy by first performing a perceptual bottom-up grouping of edge pixels to generate salient contour curves. A subsequent recursive top-down grouping yields a hierarchy of compositions. All entities in the compositional hierarchy are incorporated in a Bayesian network that couples them together by means of a shape model. The probabilistic model underlying top-down grouping as well as the shape model is learned automatically from a set of training images for the given categories. As a consequence, compositionality simplifies the learning of complex category models by building them from simple, frequently used compositions. The architecture is evaluated on the highly challenging Caltech 101 database which exhibits large intra-category variations. The proposed compositional approach shows competitive retrieval rates in the range of 53.0 ± 0.49%.
BibTeX:
@inproceedings{Ommer2006,
  author = {Bj\"orn Ommer and Michael Sauter and Joachim M. Buhmann},
  title = {Learning Top-Down Grouping of Compositional Hierarchies for Recognition},
  booktitle = {CVPR'06 (POCV)},
  year = {2006}
}
Orbanz, P. & Buhmann, J. M. Smooth Image Segmentation by Nonparametric Bayesian Inference 2006 European Conference on Computer Vision (ECCV 2006)   inproceedings  
BibTeX:
@inproceedings{Orbanz2006,
  author = {Peter Orbanz and Joachim M. Buhmann},
  title = {Smooth Image Segmentation by Nonparametric Bayesian Inference},
  booktitle = {European Conference on Computer Vision (ECCV 2006)},
  publisher = {Springer Verlag},
  year = {2006},
  volume = {I},
  pages = {444-457}
}
Peter, H., Fischer, B. & Buhmann, J. M. Probabilistic De Novo Peptide Sequencing with Doubly Charged Ions 2006 Pattern Recognition - Symposium of the DAGM 2006   inproceedings URL [PDF]  
Abstract: Sequencing of peptides by tandem mass spectrometry has matured to the key technology for proteomics. Noise in the measurement process strongly favors statistical models like NovoHMM, a recently published generative approach based on factorial hidden Markov models. We extend this hidden Markov model to include information of doubly charged ions since the original model can only cope with singly charged ions. This modification requires a refined discretization of the mass scale and, thereby, it increases its sensitivity and recall performance on a number of datasets to compare favorably with alternative approaches for mass spectra interpretation.
BibTeX:
@inproceedings{peter.fischer.ea:probabilistic,
  author = {Hansruedi Peter and Bernd Fischer and Joachim M. Buhmann},
  title = {Probabilistic De Novo Peptide Sequencing with Doubly Charged Ions},
  booktitle = {Pattern Recognition - Symposium of the DAGM 2006},
  publisher = {Springer},
  year = {2006},
  volume = {4174},
  pages = {424--433},
  url = {http://dx.doi.org/10.1007/11861898_43}
}
Rabinovich, A., Lange, T., Buhmann, J. & Belongie, S. Model Order Selection and Cue Combination for Image Segmentation 2006 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2006)   inproceedings  
BibTeX:
@inproceedings{Rabinovich2006,
  author = {A. Rabinovich and T. Lange and J. Buhmann and S. Belongie},
  title = {Model Order Selection and Cue Combination for Image Segmentation},
  booktitle = {2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2006)},
  publisher = {IEEE Computer Society},
  year = {2006},
  pages = {1130-1137}
}
Roth, V. Kernel Fisher Discriminants for Outlier Detection 2006 Neural Computation   article [PDF]  
Abstract: he problem of detecting ``atypical objects'' or ``outliers'' is one of the classical topics in (robust) statistics. Recently, it has been proposed to address this problem by means of one-class SVM classifiers. The method presented in this paper bridges the gap between kernelized one-class classification and Gaussian density estimation in the induced feature space. Having established the exact relation between the two concepts, it is now possible to identify ``atypical objects'' by quantifying their deviations from the Gaussian model. This model-based formalization of outliers overcomes the main conceptual shortcoming of most one-class approaches which, in a strict sense, are unable to detect outliers, since the expected fraction of outliers has to be specified in advance. In order to overcome the inherent model selection problem of unsupervised kernel methods, a cross-validated likelihood criterion for selecting all free model parameters is applied. Experiments for detecting atypical objects in image databases effectively demonstrate the applicability of the proposed method in real world scenarios.
BibTeX:
@article{Roth2006,
  author = {Volker Roth},
  title = {Kernel Fisher Discriminants for Outlier Detection},
  journal = {Neural Computation},
  year = {2006},
  volume = {18},
  number = {4}
}
Roth, V. & Fischer, B. Improved Functional Prediction of Proteins by Learning Kernel Combinations in Multilabel Settings 2006   techreport [PDF]  
Abstract: A method for learning kernel combinations is presented which explicitly addresses the problem of multilabel classification in which an object can belong to more than one class. Such multilabels appear naturally in the context of protein function prediction, since a protein can be involved in several different biological processes exhibiting more than one function.
BibTeX:
@techreport{Roth2006b,
  author = {Volker Roth and Bernd Fischer},
  title = {Improved Functional Prediction of Proteins by Learning Kernel Combinations in Multilabel Settings},
  year = {2006},
  number = {B-2006-4},
  note = {Proceeding of PMSB 2006}
}
Roth, V. & Ommer, B. Exploiting Low-level Image Segmentation for Object Recognition 2006 Pattern Recognition--DAGM 2006   inproceedings [PDF]  
Abstract: A method for exploiting the information in low-level image segmentations for the purpose of object recognition is presented. The key idea is to use a whole ensemble of segmentations per image, computed on different random samples of image sites. Along the boundaries of those segmentations that are stable under the sampling process we extract strings of vectors that contain local image descriptors like shape, texture and intensities. Pairs of such strings are aligned, and based on the alignment scores a mixture model is trained which divides the segments in an image into fore- and background. Given such candidate foreground segments, we show that it is possible to build a state-of-the-art object recognition system that exhibits excellent performance on a standard benchmark database. This result shows that despite the inherent problems of low-level image segmentation in poor data conditions, segmentation can indeed be a valuable tool for object recognition in real-world images.
BibTeX:
@inproceedings{Roth2006a,
  author = {Volker Roth and Bj{\"o}rn Ommer},
  title = {Exploiting Low-level Image Segmentation for Object Recognition},
  booktitle = {Pattern Recognition--DAGM 2006},
  publisher = {Springer},
  year = {2006},
  volume = {4174}
}
Steinhage, V., Schröder, S., Roth, V., Cremers, A., Drescher, W. & Wittmann, D. The Sience of Fingerprinting Bees 2006 German Research   article [PDF]  
Abstract: The ABIS system allows the rapid and accurate identification of bee species. This process is based on automatic image analysis of the bee's wings and enables the development of new, computer-aided methods to identify the bee species.
BibTeX:
@article{Steinhage2006,
  author = {V. Steinhage and S. Schr{\"o}der and V. Roth and A.B. Cremers and W. Drescher and D. Wittmann},
  title = {The Sience of Fingerprinting Bees},
  journal = {German Research},
  year = {2006},
  volume = {28},
  number = {1},
  pages = {19-21}
}
Wey, P., Fischer, B., Bay, H. & Buhmann, J. M. Dense Stereo by Triangular Meshing and Cross Validation 2006 Pattern Recognition - Symposium of the DAGM 2006   inproceedings URL [PDF]  
Abstract: Dense depth maps can be estimated in a Bayesian sense from multiple calibrated still images of a rigid scene relative to a reference view. This well-established probabilistic framework is extended by adaptively refining a triangular meshing procedure and by automatic cross-validation of model parameters. The adaptive refinement strategy locally adjusts the triangular meshing according to the measured image data. The new method substantially outperforms the competing techniques both in terms of robustness and accuracy.
BibTeX:
@inproceedings{wey.fischer.ea:dense,
  author = {Peter Wey and Bernd Fischer and Herbert Bay and Joachim M. Buhmann},
  title = {Dense Stereo by Triangular Meshing and Cross Validation},
  booktitle = {Pattern Recognition - Symposium of the DAGM 2006},
  publisher = {Springer},
  year = {2006},
  volume = {4174},
  pages = {708--717},
  url = {http://dx.doi.org/10.1007/11861898_71}
}
Buhmann, J. M., Lange, T. & Ramacher, U. Image Segmentation by Networks of Spiking Neurons 2005 Neural Computation   article  
BibTeX:
@article{Buhmann2005,
  author = {Joachim M. Buhmann and Tilman Lange and Ulrich Ramacher},
  title = {Image Segmentation by Networks of Spiking Neurons},
  journal = {Neural Computation},
  year = {2005},
  volume = {17},
  number = {5},
  pages = {1010-1031}
}
Fischer, B., Roth, V., Buhmann, J. M., Grossmann, J., Baginsky, S., Gruissem, W., Roos, F. & Widmayer, P. A Hidden Markov Model for de Novo Peptide Sequencing 2005 Neural Information Processing Systems   article URL [PDF]  
Abstract: De novo Sequencing of peptides is a challenging task in proteome research. While there exist reliable DNA-sequencing methods, the highthroughput de novo sequencing of proteins by mass spectrometry is still an open problem. Current approaches suffer from a lack in precision to detect mass peaks in the spectrograms. In this paper we present a novel method for de novo peptide sequencing based on a hidden Markov model. Experiments effectively demonstrate that this new method significantly outperforms standard approaches in matching quality.
BibTeX:
@article{fischer.roth.ea:hidden,
  author = {Bernd Fischer and Volker Roth and Joachim M. Buhmann and Jonas Grossmann and Sacha Baginsky and Wilhelm Gruissem and Franz Roos and Peter Widmayer},
  title = {A Hidden Markov Model for de Novo Peptide Sequencing},
  journal = {Neural Information Processing Systems},
  year = {2005},
  volume = {17},
  pages = {457--464},
  url = {http://books.nips.cc/papers/files/nips17/NIPS2004_0372.pdf}
}
Fischer, B., Roth, V., Roos, F., Grossmann, J., Baginsky, S., Widmayer, P., Gruissem, W. & Buhmann, J. M. NovoHMM: A Hidden Markov Model for de Novo Peptide Sequencing 2005 Analytical Chemistry   article URL [PDF]  
Abstract: De novo sequencing of peptides poses one of the most challenging tasks in data analysis for proteome research. In this paper, a generative hidden Markov model (HMM) of mass spectra for de novo peptide sequencing which constitutes a novel view on how to solve this problem in a Bayesian framework is proposed. Further extensions of the model structure to a graphical model and a factorial HMM to substantially improve the peptide identification results are demonstrated. Inference with the graphical model for de novo peptide sequencing estimates posterior probabilities for amino acids rather than scores for single symbols in the sequence. Our model outperforms state-of-the-art methods for de novo peptide sequencing on a large test set of spectra.
BibTeX:
@article{fischer.roth.ea:novohmm,
  author = {Bernd Fischer and Volker Roth and Franz Roos and Jonas Grossmann and Sacha Baginsky and Peter Widmayer and Wilhelm Gruissem and Joachim M. Buhmann},
  title = {NovoHMM: A Hidden Markov Model for de Novo Peptide Sequencing},
  journal = {Analytical Chemistry},
  year = {2005},
  volume = {77},
  number = {22},
  pages = {7265-7273},
  url = {http://dx.doi.org/10.1021/ac0508853}
}
Heiler, M., Keuchel, J. & Schnörr, C. Semidefinite Clustering for Image Segmentation with A-priori Knowledge 2005 Pattern Recognition (27th DAGM Symposium)   inproceedings  
BibTeX:
@inproceedings{Heiler2005,
  author = {M. Heiler and J. Keuchel and C. Schn{\"o}rr},
  title = {Semidefinite Clustering for Image Segmentation with A-priori Knowledge},
  booktitle = {Pattern Recognition (27th DAGM Symposium)},
  publisher = {Springer},
  year = {2005},
  volume = {3663},
  pages = {309-317}
}
Lange, T. & Buhmann, J. Combining partitions by probabilistic label aggregation 2005 KDD, Proceedings of the Eleventh ACM SIGKDD International onference on Knowledge Discovery and Data Mining   inproceedings  
BibTeX:
@inproceedings{Lange2005a,
  author = {Tilman Lange and Joachim Buhmann},
  title = {Combining partitions by probabilistic label aggregation},
  booktitle = {KDD, Proceedings of the Eleventh ACM SIGKDD International onference on Knowledge Discovery and Data Mining},
  publisher = {ACM},
  year = {2005},
  pages = {147-156}
}
Lange, T., Law, M. H. C., Jain, A. K. & Buhmann, J. M. Learning with Constrained and Unlabelled Data 2005 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2005)   inproceedings  
BibTeX:
@inproceedings{Lange2005,
  author = {Tilman Lange and Martin H. C. Law and Anil K. Jain and Joachim M. Buhmann},
  title = {Learning with Constrained and Unlabelled Data},
  booktitle = {2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2005)},
  publisher = {IEEE Computer Society},
  year = {2005},
  pages = {731-738}
}
Ommer, B. & Buhmann, J. M. Object Categorization by Compositional Graphical Models 2005 Energy Minimization Methods in Computer Vision and Pattern Recognition, LNCS 3757   inbook [PDF]  
Abstract: This contribution proposes a compositionality architecture for visual object categorization, i.e., learning and recognizing multiple visual object classes in unsegmented, cluttered real-world scenes. We propose a sparse image representation based on localized feature histograms of salient regions. Category specific information is then aggregated by using relations from perceptual organization to form compositions of these descriptors. The underlying concept of image region aggregation to condense semantic information advocates for a statistical representation founded on graphical models. On the basis of this structure, objects and their constituent parts are localized. To complement the learned dependencies between compositions and categories, a global shape model of all compositions that form an object is trained. During inference, belief propagation reconciles bottom-up feature-driven categorization with top-down category models. The system achieves a competitive recognition performance on the standard CalTech database.
BibTeX:
@inbook{Ommer2005,
  author = {Bj\"orn Ommer and Joachim M. Buhmann},
  title = {Object Categorization by Compositional Graphical Models},
  booktitle = {Energy Minimization Methods in Computer Vision and Pattern Recognition, LNCS 3757},
  publisher = {Springer},
  year = {2005},
  pages = {235-250}
}
Orbanz, P. & Buhmann, J. M. SAR Images as Mixtures of Gaussian Mixtures 2005 Proceedings of the IEEE International Conference on Image Processing (ICIP)   inproceedings  
BibTeX:
@inproceedings{Orbanz2005,
  author = {P. Orbanz and J.~M. Buhmann},
  title = {{SAR} Images as Mixtures of {G}aussian Mixtures},
  booktitle = {Proceedings of the IEEE International Conference on Image Processing (ICIP)},
  year = {2005},
  volume = {2},
  pages = {209-212}
}
Roth, V. Outlier Detection with One-class Kernel Fisher Discriminants 2005 NIPS 17   inproceedings [PDF]  
Abstract: The problem of detecting “atypical objects” or “outliers” is one of the classical topics in (robust) statistics. Recently, it has been proposed to address this problem by means of one-class SVM classifiers. The main conceptual shortcoming of most one-class approaches, however, is that in a strict sense they are unable to detect outliers, since the expected fraction of outliers has to be specified in advance. The method presented in this paper overcomes this problem by relating kernelized one-class classification to Gaussian density estimation in the induced feature space. Having established this relation, it is possible to identify “atypical objects” by quantifying their deviations from the Gaussian model. For RBF kernels it is shown that the Gaussian model is “rich enough” in the sense that it asymptotically provides an unbiased estimator for the true density. In order to overcome the inherent model selection problem, a cross-validated likelihood criterion for selecting all free model parameters is applied.
BibTeX:
@inproceedings{Roth2005,
  author = {Volker Roth},
  title = {Outlier Detection with One-class Kernel Fisher Discriminants},
  booktitle = {NIPS 17},
  year = {2005}
}
Fischer, B., Roth, V. & Buhmann, J. M. Clustering with the Connectivity Kernel 2004 Neural Information Processing Systems   inproceedings URL [PDF]  
Abstract: Clustering aims at extracting hidden structure in dataset. While the problem of finding compact clusters has been widely studied in the literature, extracting arbitrarily formed elongated structures is considered a much harder problem. In this paper we present a novel clustering algorithm which tackles the problem by a two step procedure: first the data are transformed in such a way that elongated structures become compact ones. In a second step, these new objects are clustered by optimizing a compactness-based criterion. The advantages of the method over related approaches are threefold: (i) robustness properties of compactness-based criteria naturally transfer to the problem of extracting elongated structures, leading to a model which is highly robust against outlier objects; (ii) the transformed distances induce a Mercer kernel which allows us to formulate a polynomial approximation scheme to the generally NP-hard clustering problem; (iii) the new method does not contain free kernel parameters in contrast to methods like spectral clustering or mean-shift clustering.
BibTeX:
@inproceedings{fischer.roth.ea:clustering,
  author = {Bernd Fischer and Volker Roth and Joachim M. Buhmann},
  title = {Clustering with the Connectivity Kernel},
  booktitle = {Neural Information Processing Systems},
  publisher = {MIT Press},
  year = {2004},
  volume = {16},
  url = {http://books.nips.cc/papers/files/nips16/NIPS2003_AA12.pdf}
}
Hermes, L. & Buhmann, J. Boundary-Constrained Agglomerative Segmantation 2004 IEEE Transactions on Geoscience and Remote Sensing   inproceedings [PDF]  
Abstract: Automated interpretation of remotely sensed dataposes certain demands to image segmentation algorithms, regarding speed, memory requirements, segmentation quality, noise robustness, complexity, and reproducibility. This paper addresses these issues by formulating image segmentation as source channel coding with side information. A cost function is developed thatapproximates the expected code length for a hypothetical two-part coding scheme. The cost function combines region-based and edge-based considerations, and it supports the utilization of reference data to enhance segmentation results. Optimization is implemented by an agglomerative segmentation algorithm that iteratively creates a tree-like description of the image. Given a fixed tree level and the output of the edge detector, the cost functionis parameter-free, so that no exhaustive parameter-tuning is necessary. Additionally, a criterion is presented to reliably select an adequate tree level with high descriptive quality. It is shown by statistical analysis that the cost function is appropriate for both multispectral and synthetic aperture radar data. Experimental results confirm the high quality of the resulting segmentations.
BibTeX:
@inproceedings{Hermes2004,
  author = {Lothar Hermes and Joachim Buhmann},
  title = {Boundary-Constrained Agglomerative Segmantation},
  booktitle = {IEEE Transactions on Geoscience and Remote Sensing},
  publisher = {IEEE},
  year = {2004},
  volume = {42}
}
Jain, A. K., Topchy, A., Law, M. H. & Buhmann, J. M. Landscape of Clustering Algorithms 2004 17th International Conference on Pattern Recognition Cambridge UK   inproceedings [PDF]  
Abstract: Numerous clustering algorithms, their taxonomies and evaluation studies are available in the literature. Despite the diversity of different clustering algorithms, solutions delivered by these algorithms exhibit many commonalities. An analysis of the similarity and properties of clustering objective functions is necessary from the operational/user perspective. We revisit conventional categorization of clustering algorithms and attempt to relate them according to the partitions they produce. We empirically study the similarity of clustering solutions obtained by many traditional as well as relatively recent clustering algorithms on a number of real-world data sets. Sammon’s mapping and a complete-link clustering of the inter-clustering dissimilarity values are performed to detect a meaningful grouping of the objective functions. We find that only a small number of clustering algorithms are sufficient to represent a large spectrum of clustering criteria. For example, interesting groups of clustering algorithms are centered around the graph partitioning, linkage-based and Gaussian mixture model based algorithms.
BibTeX:
@inproceedings{Jain2004,
  author = {Anil K. Jain and Alexander Topchy and Martin H.C. Law and Joachim M. Buhmann},
  title = {Landscape of Clustering Algorithms},
  booktitle = {17th International Conference on Pattern Recognition Cambridge UK},
  publisher = {IEEE},
  year = {2004},
  pages = {I-260--I-263}
}
Lange, T., Roth, V., Braun, M. L. & Buhmann, J. M. Stability-Based Validation of Clustering Solutions 2004 Neural Computation   article [PDF]  
Abstract: Data clustering describes a set of frequently employed techniques in exploratory data analysis to extract “natural” group structure in data. Such groupings need to be validated to separate the signal in the data from spurious structure. In this context, finding an appropriate number of clustersis a particularly important model selection question.We introduce a measure of cluster stability to assess the validity of a cluster model. This stability measure quantifies the reproducibility of clustering solutionson a second sample, and it can be interpreted as a classification risk with regard to class labels produced by a clustering algorithm. The preferred number of clusters is determined by minimizing this classification risk as a function of the number of clusters. Convincing results are achieved on simulated as well as gene expression data sets. Comparisons to other methods demonstrate the competitive performance of our method and its suitability as a general validation tool for clustering solutions in realworld problems.
BibTeX:
@article{Lange2004,
  author = {Tilman Lange and Volker Roth and Mikio L. Braun and Joachim M. Buhmann},
  title = {Stability-Based Validation of Clustering Solutions},
  journal = {Neural Computation},
  year = {2004},
  volume = {16},
  pages = {1299-1323}
}
Roth, V. & Lange, T. Adaptive Feature Selection in Image Segmentation 2004 Pattern Recognition, DAGM   inproceedings [PDF]  
Abstract: Most image segmentation algorithms optimize some mathematical similarity criterion derived from several low-level image features. One possible way of combining different types of features, e.g. color- and texture features on different scales and/or different orientations, is to simply stack all the individualmeasurements into one high-dimensional feature vector. Due to the nature of such stacked vectors, however, only very few components (e.g. those which are defined on a suitable scale) will carry information that is relevant for the actual segmentation task.We present an approach to combining segmentation and adaptive feature selection that overcomes this relevance determination problem. Allfree model parameters of this method are selected by a resampling-based stability analysis. Experiments demonstrate that the built-in feature selection mechanismleads to stable and meaningful partitions of the images.
BibTeX:
@inproceedings{Roth2004,
  author = {Volker Roth and Tilman Lange},
  title = {Adaptive Feature Selection in Image Segmentation},
  booktitle = {Pattern Recognition, DAGM},
  publisher = {DAGM},
  year = {2004},
  pages = {9-17}
}
Roth, V. & Lange, T. Bayesian Class Discovery in Microarray Datasets 2004 IEEE Transactions on Biomedical Engineering   article [PDF]  
Abstract: A novel approach to class discovery in gene expressiondatasets is presented. In the context of clinical diagnosis, the central goal of class discovery algorithms is to simultaneously find putative (sub-)types of diseases and to identify informative subsets of genes with disease-type specific expression profile. Contrary to many other approaches in the literature, the method presentedimplements a wrapper strategy for feature selection, in the sense that the features are directly selected by optimizing the discriminative power of the used partitioning algorithm. The usual combinatorial problems associated with wrapper approaches are overcome by a Bayesian inference mechanism. On the technical side, wepresent an efficient optimization algorithm with guaranteed local convergence property. The only free parameter of the optimization method is selected by a resampling-based stability analysis. Experiments with Leukemia and Lymphoma datasets demonstrate that our method is able to correctly infer partitions and correspondingsubsets of genes which both are relevant in a biological sense. Moreover, the frequently observed problem of ambiguities caused by different but equally high-scoring partitions is successfully overcome by the model selection method proposed.
BibTeX:
@article{Roth2004a,
  author = {Volker Roth and Tilman Lange},
  title = {Bayesian Class Discovery in Microarray Datasets},
  journal = {IEEE Transactions on Biomedical Engineering},
  year = {2004},
  volume = {51},
  number = {5}
}
Roth, V. & Lange, T. Feature Selection in Clustering Problems 2004 NIPS   inproceedings [PDF]  
Abstract: A novel approach to combining clustering and feature selection is presented. It implements a wrapper strategy for feature selection, in the sense that the features are directly selected by optimizing the discriminative power of the used partitioning algorithm. On the technical side, wepresent an efficient optimization algorithm with guaranteed local convergence property. The only free parameter of this method is selected by a resampling-based stability analysis. Experiments with real-world datasets demonstrate that our method is able to infer both meaningful partitions and meaningful subsets of features.
BibTeX:
@inproceedings{Roth2004b,
  author = {Volker Roth and Tilman Lange},
  title = {Feature Selection in Clustering Problems},
  booktitle = {NIPS},
  publisher = {MIT Press},
  year = {2004},
  pages = {1237-1244}
}
Steck, H. & Jaakkola, T. Predictive Discretization during Model Selection 2004 Pattern Recognition   article [PDF]  
Abstract: We present an approach to discretizing multivariate continuous data while learning the structure of a graphical model. We derive a joint scoring function from the principle of predictive accuracy, which inherently ensures the optimal trade-o between goodness of t and model complexity including the number of discretization levels. Using the socalled nest grid implied by the data, our scoring function depends only on the number of data points in the various discretization levels (independent of the metric used in the continuous space). Our experiments with arti cial data as well as with gene expression data show that discretization plays a crucial role regarding the resulting network structure.
BibTeX:
@article{Steck2004,
  author = {Harald Steck and Tommi Jaakkola},
  title = {Predictive Discretization during Model Selection},
  journal = {Pattern Recognition},
  year = {2004},
  volume = {LNCS 3175},
  pages = {1-8}
}
Stilkerich, S. C. & Buhmann, J. Massively Parallel Architecture for an unsupervised Segmentation Model 2004 International Conference on Signals and Electronic Systems   inproceedings [PDF]  
Abstract: In this paper we propose the structure of a novel massively parallel system-on-chip (SoC) architecture for a state-of-the-art probabilistic image segmentation model. This probabilistic model is formulated on a regular Markovian pixel grid. The unique combination of algorithmic robustness, SoC and real-time processing capabilities provides a new type of image processing systems in realworld applications. The model and the SoC architecture are extensively tested by natural images. Some chip design examples are also included to finalize the contribution.
BibTeX:
@inproceedings{Stilkerich2004,
  author = {Stephan C. Stilkerich and Joachim Buhmann},
  title = {Massively Parallel Architecture for an unsupervised Segmentation Model},
  booktitle = {International Conference on Signals and Electronic Systems},
  publisher = {IEEE},
  year = {2004}
}
Zöller, T. & Buhmann, J. M. Shape Constrained Image Segmentation by Parametric Distributional Clustering 2004 Computer Society Conference on Computer Vision and Pattern Recognition   inproceedings [PDF]  
Abstract: The automated segmentation of images into semanticallymeaningful parts requires shape information since lowlevelfeature analysis alone often fails to reach this goal.We introduce a novel method of shape constrained imagesegmentation which is based on mixtures of feature distributions for color and texture as well as probabilistic shape knowledge. The combined approach is formulated in theframework of Bayesian statistics to account for the robustness requirement in image understanding. Experimental evidence shows that semantically meaningful segments are inferred, even when image data alone gives rise to ambiguoussegmentations.
BibTeX:
@inproceedings{Zoller2004,
  author = {Thomas Z\"oller and Joachim M. Buhmann},
  title = {Shape Constrained Image Segmentation by Parametric Distributional Clustering},
  booktitle = {Computer Society Conference on Computer Vision and Pattern Recognition},
  year = {2004},
  pages = {386-394}
}
Chen, W. & Buhmann, J. M. A New Distance Measure for Probabilistic 2003 DAGM   inproceedings [PDF]  
Abstract: The contour of a planar shape is essentially one-dimensional signal embedded in 2-D space; thus the orthogonal distance, which only considers 1-D (norm) deviation from suggested models, is not rich enoughto characterize the description quality of arbitrary model/shape pairs.This paper suggests a generalized distance measure, called Transport Distance, for probabilistic shape modeling. B-Spline primitives are used to represent models. The probability of a hypothetical model for a shape is determined on the basis of the new distance measure. Experiments show that an optimization procedure, which maximize the model probability, generates robust and visually pleasing geometric models for data.
BibTeX:
@inproceedings{Chen2003,
  author = {Wei-Jun Chen and Joachim M. Buhmann},
  title = {A New Distance Measure for Probabilistic},
  booktitle = {DAGM},
  publisher = {DAGM},
  year = {2003},
  volume = {LNCS 2781},
  pages = {507-514}
}
Fischer, B. & Buhmann, J. M. Bagging for Path-Based Clustering 2003 IEEE Transactions on Pattern Analysis and Machine Intelligence   article URL [PDF]  
Abstract: A resampling scheme for clustering with similarity to bootstrap aggregation (bagging) is presented. Bagging is used to improve the quality of path-based clustering, a data clustering method that can extract elongated structures from data in a noise robust way. The results of an agglomerative optimization method are influenced by small fluctuations of the input data. To increase the reliability of clustering solutions a stochastic resampling method is developed to infer consensus clusters. A related reliability measure allows us to estimate the number of clusters, based on the stability of an optimized cluster solution under resampling. The quality of path-based clustering with resampling is evaluated on a large image dataset of human segmentations.
BibTeX:
@article{fischer.buhmann:bagging,
  author = {Bernd Fischer and Joachim M. Buhmann},
  title = {Bagging for Path-Based Clustering},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2003},
  volume = {25},
  number = {11},
  pages = {1411-1415},
  url = {http://dx.doi.org/10.1109/TPAMI.2003.1240115}
}
Fischer, B. & Buhmann, J. M. Path-Based Clustering for Grouping Smooth Curves and Texture Segmentation 2003 IEEE Transactions on Pattern Analysis and Machine Intelligence   article URL [PDF]  
Abstract: Perceptual Grouping organizes image parts in clusters based on psychophysically plausible similarity measures. We propose a novel grouping method in this paper which stresses connectedness of image elements via mediating elements rather than favoring high mutual similarity. This grouping principle yields superior clustering results when objects are distributed on low-dimensional extended manifolds in a feature space and not as local point clouds. In addition to extracting connected structures, objects are singled out as outliers when they are too far away from any cluster structure. The objective function for this perceptual organization principle is optimized by a fast agglomerative algorithm. We report on perceptual organization experiments where small edge elements are grouped to smooth curves. The generality of the method is emphasized by results from grouping textured images with texture gradients in an unsupervised fashion.
BibTeX:
@article{fischer.buhmann:path-based,
  author = {Bernd Fischer and Joachim M. Buhmann},
  title = {Path-Based Clustering for Grouping Smooth Curves and Texture Segmentation},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2003},
  volume = {25},
  number = {4},
  pages = {513-518},
  url = {http://dx.doi.org/10.1109/TPAMI.2003.1190577}
}
Hermes, L. & Buhmann, J. A Minimum Entropy Approach to Adaptive Image Polygonization 2003 IEEE Transaction on Image Processing   inproceedings [PDF]  
Abstract: This paper introduces a novel adaptive image segmentationalgorithm which represents images by polygonal segments.The algorithm is based on an intuitive generative modelfor pixel intensities and its associated cost function which can be effectively optimized by a hierarchical triangulation algorithm. A triangular mesh is iteratively refined and reorganized to extract a compact description of the essential image structure. After analyzingfundamental convexity properties of our cost function, weadapt an information-theoretic bound to assess the statistical significance of a given triangulation step. The bound effectively defines a stopping criterion to limit the number of triangles in the mesh, thereby avoiding undesirable overfitting phenomena. It alsofacilitates the development of a multiscale variant of the triangulation algorithm, which substantially improves its computational demands. The algorithm has various applications in contextual classification, remote sensing, and visual object recognition. It is particularly suitable for the segmentation of noisy imagery.
BibTeX:
@inproceedings{Hermes2003,
  author = {Lothar Hermes and Joachim Buhmann},
  title = {A Minimum Entropy Approach to Adaptive Image Polygonization},
  booktitle = {IEEE Transaction on Image Processing},
  publisher = {IEEE},
  year = {2003},
  volume = {12}
}
Hermes, L. & Buhmann, J. M. Semi-Supervised Image Segmentation by Parametric Distributional Clustering 2003 Energy Minimization Methods in Computer Vision and Pattern Recognition   inproceedings [PDF]  
Abstract: The problem of semi-supervised image segmentation is fre-quently posed e.g. in remote sensing applications. In this setting, one aims at finding a decomposition of a given image into its constituent regions, which are typically assumed to have homogeneously distributed pixel values. In addition, it is requested that these regions can be equipped with some semantics, i.e. that they can be matched to particular land cover classes. For this purpose, class labels are provided for a small sub-set of the image data. The demand that the image segmentation re-spects those class labels implies that the segmentation algorithm should be posed as a constrained optimization problem.We extend the Parametric Distributional Clustering (PDC) algorithm to fit into this learning framework. The resulting optimization problem is solved by constrained Deterministic Annealing. The approach is illus-trated for both artificial data and real-world synthetic aperture radar (SAR) imagery.
BibTeX:
@inproceedings{Hermes2003a,
  author = {Lothar Hermes and Joachim M. Buhmann},
  title = {Semi-Supervised Image Segmentation by Parametric Distributional Clustering},
  booktitle = {Energy Minimization Methods in Computer Vision and Pattern Recognition},
  publisher = {CVPR},
  year = {2003},
  pages = {229-245}
}
Ommer, B. & Buhmann, J. M. A Compositionality Architecture for Perceptual Feature Grouping 2003 Energy Minimization Methods in Computer Vision and Pattern Recognition   inproceedings [PDF]  
Abstract: We propose a compositionality architecture for perceptual organization which establishes a novel, generic, algorithmic framework for feature binding and condensation of semantic information in im-ages. The underlying algorithmic ideas require a hierarchical structure for various types of objects and their groupings, which are guided by gestalt laws from psychology. A rich set of predefined feature detectors with uncertainty that perform real-valued measurements of relationships between objects can be combined in this exible Bayesian framework. Compositions are inferred by minimizing the negative posterior group-ing probability. The model structure is founded on the fundamental per-ceptual law of Prägnanz. The grouping algorithm performs hierarchical agglomerative clustering and it is rendered computationally feasible by visual pop-out. Evaluation on the edgel grouping task confirms the ro-bustness of the architecture and its applicability to grouping in various visual scenarios.
BibTeX:
@inproceedings{Ommer2003,
  author = {Bj\"orn Ommer and Joachim M. Buhmann},
  title = {A Compositionality Architecture for Perceptual Feature Grouping},
  booktitle = {Energy Minimization Methods in Computer Vision and Pattern Recognition},
  publisher = {CVPR},
  year = {2003},
  volume = {LNCS 2683},
  pages = {275-290}
}
Roth, V., Laub, J., Buhmann, J. M. & Müller, K. Going Metric: Denoising Pairwise Data 2003 Advances in Neutral Information Processing Systems   inproceedings [PDF]  
Abstract: Pairwise data in empirical sciences typically violate metricity, either due to noise or due to fallible estimates, and therefore are hard to analyze by conventional machine learning technology. In this paper we therefore stduy ways to work around this problem. First, we present an alternative embedding to multi-dimensional scaling (MDS) that allows us to apply a variety of classical machine learning and signal processing algorithms. The class of pairwise grouping algorithms which share the shift-invariance property is statistically invariant under this embedding procedure. leading to identical assignments of objects to clusters. Based on this new vestorial representation, denoising methods are applied in a second step. Both steps provide a theoretically well controlled setup to translate from pairwise data to the respective denoised metric representation. We demonstrate the practical usefulness of our theoretical reasoning by discovering structure in protein sequence data bases, visibly improving performance upon existing automatic methods.
BibTeX:
@inproceedings{Roth2003,
  author = {Volker Roth and Julian Laub and Joachim M. Buhmann and Klaus-Robert M\"uller},
  title = {Going Metric: Denoising Pairwise Data},
  booktitle = {Advances in Neutral Information Processing Systems},
  publisher = {NIPS},
  year = {2003},
  volume = {15},
  pages = {817-824}
}
Roth, V., Laub, J., Kawanabe, M. & Buhmann, J. M. Optimal Cluster Preserving Embedding of Nonmetric Proximity Data 2003 IEEE Transactions on Pattern Analysis and Machine Intelligence   article [PDF]  
Abstract: For several major applications of data analysis, objects are often not represented as feature vectors in a vector space, but rather by a matrix gathering pairwise proximities. Such pairwise data often violates metricity and, therefore, cannot be naturally embedded in a vector space. Concerning the problem of unsupervised structure detection or clustering, in this paper, a new embedding method for pairwise data into Euclidean vector spaces is introduced. We show that all clustering methods, which areinvariant under additive shifts of the pairwise proximities, can be reformulated as grouping problems in Euclidian spaces. The most prominent property of this constant shift embedding framework is the complete preservation of the cluster structure in the embeddingspace. Restating pairwise clustering problems in vector spaces has several important consequences, such as the statistical description of the clusters by way of cluster prototypes, the generic extension of the grouping procedure to a discriminative prediction rule, and theapplicability of standard preprocessing methods like denoising or dimensionality reduction.
BibTeX:
@article{Roth2003a,
  author = {Volker Roth and Julian Laub and Motoaki Kawanabe and Joachim M. Buhmann},
  title = {Optimal Cluster Preserving Embedding of Nonmetric Proximity Data},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2003},
  volume = {25},
  number = {12}
}
Braun, M. L. & Buhmann, J. M. The Noisy Euclidean Traveling Salesman Problem and Learning 2002 NIPS   inproceedings [PDF]  
Abstract: We consider noisy Euclidean traveling salesman problems in the plane, which are random combinatorial problems with underlying structure. Gibbs sampling is used to compute average trajectories, which estimate the underlying structure common to all instances. This procedure requires identifying the exact relationship between permutations and tours. In a learning setting, the average trajectory is used as a model to construct solutions to new instances sampledfrom the same source. Experimental results show that the average trajectory can in fact estimate the underlying structure and that overfitting effects occur if the trajectory adapts too closely to a single instance.
BibTeX:
@inproceedings{Braun2002,
  author = {Mikio L. Braun and Joachim M. Buhmann},
  title = {The Noisy Euclidean Traveling Salesman Problem and Learning},
  booktitle = {NIPS},
  publisher = {NIPS},
  year = {2002},
  volume = {14}
}
Buhmann, J. M. Data Clustering and Learning 2002 Handbook of Brain Theory and Neural Networks   inbook [PDF]  
Abstract: Intelligent data analysis extracts symbolic information and relations between objects from quantitative or qualitative data. A prominent class of methods are clustering or grouping principles which are designed to discover and extract structures hidden in data sets. The parameters which represent the clusters are either estimated on the basis of quality criteria or cost functions or, alternatively, they are derived by local search algorithms which are not necessarily following the gradient of a global quality criterion. This approach to inference of structure in data is known as unsupervised learning in the neural computation literature. Clustering as a fundamental pattern recognition problem can be characterized by the following design steps:1) Data Representation: What data types represent the objects in the best way to stress relations between the objects, e.g., similarity? 2) Modeling: How can we formally characterize interesting and relevant cluster structures in data sets? 3) Optimization: How can we e ciently search for cluster structures? 4) Validation: How can we validate selected or learned structures?It is important to note that the data representation issue predetermines what kind of cluster structures can be discovered in the data. Vectorial data, proximity or similarity data and histogram data are three examples of a wide variety of data types which are analyzed in the clustering literature. On the basis of the data representation, the modeling of clusters de nes the notion of groups of clusters in the data and separates desired group structures from unfavorable ones. We consider it mandatory that the modeling step yields a quality measure which is either optimized or approximated during the search for hidden structure in data sets. Formulating the search for clusters as an optimization problem allows us to validate clustering results by large deviation estimates, i.e., robust cluster structures in data should vary little from one data set to a second data set generated by the same data source.
BibTeX:
@inbook{Buhmann2002,
  author = {Joachim M. Buhmann},
  title = {Data Clustering and Learning},
  booktitle = {Handbook of Brain Theory and Neural Networks},
  year = {2002},
  pages = {333-333},
  edition = {2nd}
}
Fischer, B. & Buhmann, J. M. Data Resampling for Path Based Clustering 2002 Pattern Recognition - Symposium of the DAGM 2002   inproceedings URL [PDF]  
Abstract: Path Based Clustering assigns two objects to the same cluster if they are connected by a path with high similarity between adjacent objects on the path. In this paper, we propose a fast agglomerative algorithm to minimize the Path Based Clustering cost function. To enhance the reliability of the clustering results a stochastic resampling method is used to generate candidate solutions which are merged to yield empirical assignment probabilities of objects to clusters. The resampling algorithm measures the reliability of the clustering solution and, based on their stability, determines the number of clusters.
BibTeX:
@inproceedings{fischer.buhmann:data,
  author = {Bernd Fischer and Joachim M. Buhmann},
  title = {Data Resampling for Path Based Clustering},
  booktitle = {Pattern Recognition - Symposium of the DAGM 2002},
  year = {2002},
  volume = {LNCS 2449},
  pages = {206 - 214},
  url = {http://www.springerlink.com/content/pubpn9pgcbmrgtxc/}
}
Hermes, L., Zöller, T. & Buhmann, J. M. Parametric Distributional Clustering for Image Segmentation 2002 Computer Vision - ECCV 2002   inproceedings [PDF]  
Abstract: Unsupervised Image Segmentation is one of the central issues in Computer Vision. From the viewpoint of exploratory data analysis, segmentation can be formulated as a clustering problem in which pixels or small image patches are grouped together based on local feature information. In this contribution, parametrical distributional clustering (PDC) is presented as a novel approach to image segmentation. In contrast to noise sensitive point measurements, local distributions of image features provide a statistically robust description of the local image properties. The segmentation technique is formulated as a generative model in the maximum likelihood framework. Moreover, there exists an insightful connection to the novel information theoretic concept of the Information Bottleneck (Tishby et al., 1999), which emphasizes the compromise between efficient coding of an image and preservation of characteristic information in the measured feature distributions.The search for good grouping solutions is posed as an optimization problem, which is solved by deterministic annealing techniques. In order to further increase the computational efficiency of the resulting segmentation algorithm, a multi-scale optimization scheme is developed. Finally, the performance of the novel model is demonstrated by segmentation of color images from the Corel data base.
BibTeX:
@inproceedings{Hermes2002,
  author = {Lothar Hermes and Thomas Z\"oller and Joachim M. Buhmann},
  title = {Parametric Distributional Clustering for Image Segmentation},
  booktitle = {Computer Vision - ECCV 2002},
  publisher = {A. Heyden, G. Sparr, M. Nielsen, P. Johansen},
  year = {2002},
  volume = {3},
  pages = {577-591}
}
Lange, T., Braun, M., Roth, V. & Buhmann, J. Stability-Based Model Selection 2002 NIPS   inproceedings [PDF]  
Abstract: Model selection is linked to model assessment, which is the problem of comparing different models, or model parameters, for a specific learning task. For supervised learning, the standard practical technique is cross-validation, which is not applicable for semi-supervised and unsupervised settings. In this paper, a new model assessment scheme is introduced which is based on a notion of stability. The stability measure yields an upper bound to cross-validation in the supervised case, but extends to semi-supervised and unsupervised problems. In the experimental part, the performance of the stability measure is studied for model order selection in comparison to standard techniques in this area.
BibTeX:
@inproceedings{Lange2002,
  author = {Tilman Lange and Mikio Braun and Volker Roth and Joachim Buhmann},
  title = {Stability-Based Model Selection},
  booktitle = {NIPS },
  publisher = {NIPS },
  year = {2002}
}
Marx, Z., Dagan, I., Buhmann, J. M. & Shamir, E. Coupled Clustering: A Method for Detecting Structural Correspondence 2002 Journal of Machine Learning Research   article [PDF]  
Abstract: This paper proposes a new paradigm and a computational framework for revealing equivalencies (analogies) between sub-structures of distinct composite systems that are initially represented by unstructured data sets. For this purpose, we introduce and investigate a variant of traditional data clustering, termed coupled clustering, which outputs a configuration of corresponding subsets of two such representative sets. We apply our method to synthetic as well as textual data. Its achievements in detecting topical correspondences between textual corpora are evaluated through comparison to performance of human experts.
BibTeX:
@article{Marx2002,
  author = {Zvika Marx and Ido Dagan and Joachim M. Buhmann and Eli Shamir},
  title = {Coupled Clustering: A Method for Detecting Structural Correspondence},
  journal = {Journal of Machine Learning Research},
  year = {2002},
  volume = {3},
  pages = {747-780}
}
Roth, V., Braun, M., Lange, T. & Buhmann, J. M. Stability-Based Model Order Selection in Clustering with Applications to Gene Expression Data 2002 Artificial Neural Networks - ICANN 2002   inproceedings [PDF]  
Abstract: The concept of cluster stability is introduced as a means for assessing the validity of data partitionings found by clustering algorithms. It allows us to explicitly quantify the quality of a clustering solution, without being dependent on external information. The principle of maximizing the cluster stability can be interpreted as choosing the most self-consistent data partitioning. We present an empirical estimator for the theoretically derived stability index, based on resampling. Experiments are conducted on well known gene expression data sets, re-analyzing the work by Alon et al. [1] and by Spellman et al. [2].
BibTeX:
@inproceedings{Roth2002a,
  author = {Volker Roth and Mikio Braun and Tilman Lange and Joachim M. Buhmann},
  title = {Stability-Based Model Order Selection in Clustering with Applications to Gene Expression Data},
  booktitle = {Artificial Neural Networks - ICANN 2002},
  year = {2002},
  volume = {LNCS 2415}
}
Roth, V., Lange, T., Braun, M. & Buhmann, J. A Resampling Approach to Cluster Validation 2002 15th Symposium Held in Berlin , Germany 2002 (COMPSTAT2002)   inproceedings [PDF]  
Abstract: The concept of cluster stability is introduced as a means for assessing the validity of data partitionings found by clustering algorithms. It allows us to explicitly quantify the quality of a clustering solution, without being dependent on external information. The principle of maximizing the cluster stability can be interpreted as choosing the most self-consistent data partitioning. We present an empirical estimator for the theoretically derived stability index, based on imitating independent sample-sets by way of resampling. Experiments on both toy-examples and real-world problems effectively demonstrate that the proposed validation principle is highly suited for model selection.
BibTeX:
@inproceedings{Roth2002,
  author = {Volker Roth and Tilman Lange and Mikio Braun and Joachim Buhmann},
  title = {A Resampling Approach to Cluster Validation},
  booktitle = {15th Symposium Held in Berlin , Germany 2002 (COMPSTAT2002)},
  year = {2002},
  pages = {123-128}
}
Zöller, T. & Buhmann, J. M. Self-organized Clustering of Mixture Models for Combined Color and Texture Segmentation 2002 The 2nd international workshop on texture analysis and segmentation   inproceedings [PDF]  
Abstract: The segmentation of images based on color and texture cues is formulated as a clustering problem in the joint color and texture space. Small image patches are grouped together on the basis of local color and spatial frequency statistics which is captured by Gaussian mixture models in feature space. The locality of segments in feature space is taken into account by the topological organization of the cluster structure which corresponds to self-organizing maps. The clusters define a one dimensional chain in the space of mixture models, which favors an ordering of the groups with similar color and texture distributions. The probabilistic model is tested by segmentation experiments on images from the Corel data base.
BibTeX:
@inproceedings{Zoller2002a,
  author = {Thomas Z\"oller and Joachim M. Buhmann},
  title = {Self-organized Clustering of Mixture Models for Combined Color and Texture Segmentation},
  booktitle = {The 2nd international workshop on texture analysis and segmentation},
  year = {2002},
  pages = {163-167}
}
Zöller, T., Hermes, L. & Buhmann, J. M. Combined Color and Texture Segmentation by Parametric Distributional Clustering 2002 ICPR '02   inproceedings [PDF]  
Abstract: Unsupervised image segmentation can be formulated as a clustering problem in which pixels or small image patches are grouped together based on local feature information. In this contribution, parametric distributional clustering (PDC) is presented as a novel approach to image segmentation based on color and texture clues. The objective function of the PDC model is derived from the recently proposed Information Bottleneck framework (Tishby et al. [8]), but it can equivalently be formulated in terms of a maximum likelihood solution. Its optimization is performed by determin-istic annealing. Segmentation results are shown for natural wildlife imagery.
BibTeX:
@inproceedings{Zoller2002,
  author = {Thomas Z\"oller and Lothar Hermes and Joachim M. Buhmann},
  title = {Combined Color and Texture Segmentation by Parametric Distributional Clustering},
  booktitle = {ICPR '02},
  year = {2002},
  volume = {2},
  pages = {627-630}
}
Buhmann, J. M. Clustering Principles and Empirical Risk Approximation 2001 International Conference on Applied Statistical Models for Data Analysis (ASMDA'01)   inproceedings [PDF]  
Abstract: Data Clustering is one of the fundamental techniques in pattern recognition to extract structure from data. We discuss a maximum entropy approach to clustering for different clustering cost functions and relate it to the estimation principle of Empirical Risk Approximation. Large deviation techniques from statistical learning theory provide guarantees for the stability of clustering solutions.
BibTeX:
@inproceedings{Buhmann2001,
  author = {Joachim M. Buhmann},
  title = {Clustering Principles and Empirical Risk Approximation},
  booktitle = {International Conference on Applied Statistical Models for Data Analysis (ASMDA'01)},
  year = {2001}
}
Fischer, B., Zöller, T. & Buhmann, J. M. Path Based Pairwise Data Clustering with Application to Texture Segmentation 2001 Energy Minimization Methods in Computer Vision and Pattern Recognition   inproceedings URL [PDF]  
Abstract: Most cost function based clustering or partitioning methods measure the compactness of groups of data. In contrast to this picture of a point source in feature space, some data sources are spread out on a low-dimensional manifold which is embedded in a high dimensional data space. This property is adequately captured by the criterion of connectedness which is approximated by graph theoretic partitioning methods. We propose in this paper a pairwise clustering cost function with a novel dissimilarity measure emphasizing connectedness in feature space rather than compactness. The connectedness criterion considers two objects as similar if there exists a mediating intra cluster path without an edge with large cost. The cost function is optimized in a multi-scale fashion. This new path based clustering concept is applied to segment textured images with strong texture gradients based on dissimilarities between image patches.
BibTeX:
@inproceedings{fischer.zoller.ea:path,
  author = {Bernd Fischer and Thomas Z\"oller and Joachim M. Buhmann},
  title = {Path Based Pairwise Data Clustering with Application to Texture Segmentation},
  booktitle = {Energy Minimization Methods in Computer Vision and Pattern Recognition},
  year = {2001},
  volume = {LNCS 2134},
  pages = {235-250},
  url = {http://www.springerlink.com/content/a71hve79rpm83jrn/}
}
Hermes, L. & Buhmann, J. M. A New Adaptive Algorithm for the Polygonization of Noisy Imagery 2001   techreport [PDF]  
Abstract: This report introduces a novel adaptive image segmentation algorithm which represents images by polygonal segments and which is particularly suitable for the segmentation of noisy imagery. At first we suggest an intuitive generative model and its associated cost function. The cost function can effectively be optimized by a hierarchical triangulation algorithm, which iteratively refines and reorganizes a triangular mesh and finally provides a compact description of the essential image structure. After analyzing fundamental properties of our cost function, we adapt an information-theoretic bound to assess the statistical significance of a given triangulation step. The bound effectively defines a stopping criterion to limit the number of triangles in the mesh, thereby avoiding undesirable overfitting phenomena. Besides, it facilitates the development of a multi-scale variant of the triangulation algorithm, which substantially decreases its computational demands. The algorithm has various important applications in contextual classification, remote sensing, and visual object recognition.
BibTeX:
@techreport{Hermes2001,
  author = {Lothar Hermes and Joachim M. Buhmann},
  title = {A New Adaptive Algorithm for the Polygonization of Noisy Imagery},
  year = {2001},
  number = {IAI-TR-2001-3}
}
Hermes, L. & Buhmann, J. M. Contextual Classification by Entropy-Based Polygonization 2001 IEEE Conference on Computer Vision and Pattern Recognition (CVPR'01)   inproceedings [PDF]  
Abstract: To improve the performance of pixel-wise classification results for remotely sensed imagery, several contextual classification schemes have been proposed that aim at avoiding classification noise by local averaging. These algorithms, however, bear the serious disadvantage of smoothing the segment boundaries and producing rounded segments that hardly match the true shapes. In this contribution, we present a novel contextual classification algorithm that overcomes these shortcomings. Using a hierarchical approach for generating a triangular mesh, it decomposes the image into a set of polygons that, in our application, represent individual land-cover types. Compared to classical contextual classification approaches, this method has the advantage of generating output that matches the intuitively expected type of segmentation. Besides, it achieves excellent classification results.
BibTeX:
@inproceedings{Hermes2001a,
  author = {Lothar Hermes and Joachim M. Buhmann},
  title = {Contextual Classification by Entropy-Based Polygonization},
  booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR'01)},
  year = {2001},
  volume = {2},
  pages = {442-447}
}
Marx, Z., Dagan, I. & Buhmann, J. M. Coupled Clustering: A Method for Detecting Structural Correspondence 2001 18th International Conference on Machine Learning (ICML 2001)   inproceedings [PDF]  
Abstract: This paper proposes a new paradigm and a computational framework for revealing equivalencies (analogies) between sub-structures of distinct composite systems that are initially represented by unstructured data sets. For this purpose, we introduce and investigate a variant of traditional data clustering, termed coupled clustering, which outputs a configuration of corresponding subsets of two such representative sets. We apply our method to synthetic as well as textual data. Its achievements in detecting topical correspondences between textual corpora are evaluated through comparison to performance of human experts.
BibTeX:
@inproceedings{Marx2001,
  author = {Zvika Marx and Ido Dagan and Joachim M. Buhmann},
  title = {Coupled Clustering: A Method for Detecting Structural Correspondence},
  booktitle = {18th International Conference on Machine Learning (ICML 2001)},
  year = {2001},
  pages = {353-360}
}
Roth, V. Probabilistic Discriminative Kernel Classifiers for Multiclass Problems 2001 Pattern Recognition--DAGM'01   inproceedings [PDF]  
Abstract: Logistic regression is presumably the most popular representative of probabilistic discriminative classifiers. In this paper, a kernel variant of logistic regression is introduced as an iteratively re-weighted least-squares algorithm in kernel-induced feature spaces. This formulation allows us to apply highly efficient approximation methods that are capable of dealing with large-scale problems. For multi-class problems, a pairwise coupling procedure is proposed. Pairwise coupling for kernelized logistic regression effectively overcomes conceptual and numerical problems of standard multi-class kernel classifiers.
BibTeX:
@inproceedings{Roth2001a,
  author = {Volker Roth},
  title = {Probabilistic Discriminative Kernel Classifiers for Multiclass Problems},
  booktitle = {Pattern Recognition--DAGM'01},
  year = {2001},
  volume = {LNCS 2191},
  pages = {246-253}
}
Roth, V. Sparse Kernel Regressors 2001 Artificial Neural Networks--ICANN 2001   inproceedings [PDF]  
Abstract: Sparse kernel regressors have become popular by applying the support vector method to regression problems. Although this approach has been shown to exhibit excellent generalization properties in many experiments, it suffers from several drawbacks: the absence of probabilistic outputs, the restriction to Mercer kernels, and the steep growth of the number of support vectors with increasing size of the training set. In this paper we present a new class of kernel regressors that effectively overcome the above problems. We call this new approach generalized LASSO regression. It has a clear probabilistic interpretation, produces extremely sparse solutions, can handle learning sets that are corrupted by outliers, and is capable of dealing with large-scale problems.
BibTeX:
@inproceedings{Roth2001b,
  author = {Volker Roth},
  title = {Sparse Kernel Regressors},
  booktitle = {Artificial Neural Networks--ICANN 2001},
  year = {2001},
  volume = {LNCS 2130},
  pages = {339-346}
}
Roth, V. & Tsuda, K. Pairwise Coupling for Machine Recognition of Hand-Printed Japanese Characters 2001 IEEE Conference on Computer Vision and Pattern Recognition (CVPR'01)   inproceedings [PDF]  
Abstract: Machine recognition of hand-printed Japanese characters has been an area of great interest for many years. The major problem with this classification task is the huge number of different characters. Applying standard state-of-the-art techniques, such as the SVM, to multi-class problems of this kind imposes severe problems, both of a conceptual and a technical nature: (i) separating one class from all others may be an unnecessarily hard problem; (ii) solving these subproblems can impose unacceptably high computational costs. In this paper, a new approach to Japanese character recognition is presented that successfully overcomes these shortcomings. It is based on a pairwise coupling procedure for probabilistic two-class kernel classifiers. Experimental results for Hiragana recognition effectively demonstrate that our method attains an excellent level of prediction accuracy while imposing very low computational costs.
BibTeX:
@inproceedings{Roth2001,
  author = {Volker Roth and Koji Tsuda},
  title = {Pairwise Coupling for Machine Recognition of Hand-Printed Japanese Characters},
  booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR'01)},
  year = {2001},
  volume = {1},
  pages = {1120-1125}
}
Rubner, Y., Puzicha, J., Tomasi, C. & Buhmann, J. M. Empirical Evaluation of Dissimilarity Measures for Color and Texture 2001 Computer Vision and Image Understanding   article  
BibTeX:
@article{Rubner2001,
  author = {Yossi Rubner and Jan Puzicha and Carlo Tomasi and Joachim M. Buhmann},
  title = {Empirical Evaluation of Dissimilarity Measures for Color and Texture},
  journal = {Computer Vision and Image Understanding},
  year = {2001},
  volume = {84},
  pages = {25-43}
}
Stenger, B., Ramesh, V., Paragios, N., F.Coetzee & Buhmann, J. M. Topology Free Hidden Markov Models: Application to Background Modeling 2001 Proceedings International Conference on Computer Vision (ICCV 2001)   inproceedings [PDF]  
Abstract: Hidden Markov Models (HMMs) are increasingly being used in computer vision for applications such as: gesture analysis, action recognition from video, and illumination modeling. Their use involves an off-line learning step that is used as a basis for on-line decision making (i.e. a stationarity assumption on the model parameters). But, real-world applications are often non-stationary in nature. This leads to the need for a dynamic mechanism to learn and update the model topology as well as its parameters. This paper presents a new framework for HMM topology and parameter estimation in an online, dynamic fashion. The topology and parameter estimation is posed as a model selection problem with an MDL prior. Online modifications to the topology are made possible by incorporating a state splitting criterion. To demonstrate the potential of the algorithm, the background modeling problem is considered. Theoretical validation and real experiments are presented.
BibTeX:
@inproceedings{Stenger2001,
  author = {B. Stenger and V. Ramesh and N. Paragios and F.Coetzee and J. M. Buhmann},
  title = {Topology Free Hidden Markov Models: Application to Background Modeling},
  booktitle = {Proceedings International Conference on Computer Vision (ICCV 2001)},
  year = {2001},
  volume = {I},
  pages = {294-301}
}
Suing, M., Hermes, L. & Buhmann, J. M. A New Contour-based Approach to Object Recognition for Assembly Line Robots 2001 Pattern Recognition - 23rd DAGM Symposium   inproceedings [PDF]  
Abstract: A complete processing chain for visual object recognition is described in this paper. The system automatically detects individual objects on an assembly line, identifies their type, position, and orientation, and, thereby, forms the basis for automated object recognition and manipulation by single-arm robots. Two new ideas entered into the design of the recognition system. First we introduce a new fast and robust image segmentation algorithm that identifies objects in an unsupervised manner and describes them by a set of closed polygonal lines. Second we describe how to embed this object description into an object recognition process that classifies the objects by matching them to a given set of prototypes. Furthermore, the matching function allows us to determine the relative orientation and position of an object. Experimental results for a representative set of real-world tools demonstrate the quality and the practical applicability of our approach.
BibTeX:
@inproceedings{Suing2001,
  author = {Markus Suing and Lothar Hermes and and Joachim M. Buhmann},
  title = {A New Contour-based Approach to Object Recognition for Assembly Line Robots},
  booktitle = {Pattern Recognition - 23rd DAGM Symposium},
  year = {2001},
  volume = {LNCS 2191},
  pages = {329-336}
}
Bengio, Y., Buhmann, J. M., Embrechts, M. & Zurada, J. Introduction to the Special Issue on Neural Networks for Data Mining and Knowledge Discovery 2000 IEEE Transactions on Neural Networks   article  
BibTeX:
@article{Bengio2000,
  author = {Yoshua Bengio and Joachim M. Buhmann and Mark Embrechts and Jacek Zurada},
  title = {Introduction to the Special Issue on Neural Networks for Data Mining and Knowledge Discovery},
  journal = {IEEE Transactions on Neural Networks},
  year = {2000},
  volume = {11},
  number = {3},
  pages = {545-549}
}
Buhmann, J. M. & Held, M. Model selection in clustering by uniform convergence bounds 2000 Advances in Neural Information Processing Systems 12   inproceedings  
Abstract: Unsupervised learning algorithms are designed to extract structure from data samples. Reliable and robust inference requires a guarantee that extracted structures are typical for the data source, i.e., similar structures have to be infered from a second sample set of the same data source. The overfitting phenomenon in maximum entropy based annealing algorithms is exemplarily studied for a class of histogram clustering models. Bernstein's inequality for large deviations is used to determine the maximally achievable approximation quality parameterized by a minimal temperature. Monte Carlo simulations support the proposed model selection criterion by finite temperature annealing.
BibTeX:
@inproceedings{Buhmann2000,
  author = {Joachim M. Buhmann and Marcus Held},
  title = {Model selection in clustering by uniform convergence bounds},
  booktitle = {Advances in Neural Information Processing Systems 12},
  year = {2000},
  pages = {216-222}
}
Hermes, L. & Buhmann, J. M. Feature Selection for Support Vector Machines 2000 International Conference on Pattern Recognition (ICPR'00)   inproceedings [PDF]  
Abstract: In the context of support vector machines (SVM), high dimensional input vectors often reduce the computational efficiency and significantly slow down the classification process. In this paper, we propose a strategy to rank individual components according to their influence on the class assignments. This ranking is used to select an appropriate subset of the features. It replaces the original feature set without significant loss in classification accuracy. Often, the generalization ability of the classifier even increases due to the implicit regularization achieved by feature pruning.
BibTeX:
@inproceedings{Hermes2000,
  author = {Lothar Hermes and Joachim M. Buhmann},
  title = {Feature Selection for Support Vector Machines},
  booktitle = {International Conference on Pattern Recognition (ICPR'00)},
  year = {2000},
  volume = {2},
  pages = {716-719}
}
Klock, H. & Buhmann, J. M. Data Visualization by Multidimensional Scaling: A Deterministic Annealing Approach 2000 Pattern Recognition   article  
BibTeX:
@article{Klock2000,
  author = {Hansj\"org Klock and Joachim M. Buhmann},
  title = {Data Visualization by Multidimensional Scaling: A Deterministic Annealing Approach},
  journal = {Pattern Recognition},
  year = {2000},
  volume = {33},
  number = {4},
  pages = {651-669}
}
Puzicha, J., Held, M., Ketterer, J., Buhmann, J. M. & Fellner, D. On Spatial Quantization of Color Images 2000 IEEE Transactions in Image Processing   article [PDF]  
Abstract: Image quantization and digital halftoning are fundamental image processing problems in computer vision and graphics. Both steps are generally performed sequentially and, in most cases, independent of each other. Color reduction with a pixel-wise defined distortion measure and the halftoning process with its local averaging neighborhood typically optimize different quality criteria or, frequently, follow a heuristic approach without reference to any quality measure. In this paper we propose a new model to simultaneously quantize and halftone color images. The method is based on a rigorous cost-function approach which optimizes a quality criterion derived from a simplified model of human perception. It incorporates spatial and contextual information into the quantization and thus overcomes the artificial separation of quantization and halftoning. Optimization is performed by an efficient multiscale procedure which substantially alleviates the computational burden. The quality criterion and the optimization algorithms are evaluated on a representative set of artificial and real-world images showing a significant image quality improvement compared to standard color reduction approaches. Applying the developed cost function we also suggest a new distortion measure for evaluating the quality of color reduction schemes.
BibTeX:
@article{Puzicha2000,
  author = {Jan Puzicha and Markus Held and Jens Ketterer and Joachim M. Buhmann and Dieter Fellner},
  title = {On Spatial Quantization of Color Images},
  journal = {IEEE Transactions in Image Processing},
  year = {2000},
  volume = {9},
  number = {4},
  pages = {666-682}
}
Will, S., Hermes, L., Buhmann, J. & Puzicha, J. On Learning Texture Edge Detectors 2000 International Conference on Image Processing (ICIP'00)   inproceedings [PDF]  
Abstract: Texture is an inherently non-local image property. All common texture descriptors, therefore, have a significant spatial support which renders classical edge detection schemes inadequate for the detection of texture boundaries.In this paper we propose a novel scheme to learn filters for texture edge detection. Textures are defined by the statistical distribution of Gabor filter responses. Optimality criteria for detection reliability and localization accuracy are suggested in the spirit of Canny's edge detector. Texture edges are determined as zero crossings of the difference of the two a posteriori class distributions. An optimization algorithm is designed to determine the best filter kernel according to the underlying quality measure. The effectiveness of the approach is demonstrated on texture mondrians composed from the Brodatz album and a series of synthetic aperture radar (SAR) imagery. Moreover, we indicate how the proposed scheme can be combined with snake-type algorithms for prior-knowledge driven boundary refinement and interactive annotation.
BibTeX:
@inproceedings{Will2000,
  author = {Stefan Will and Lothar Hermes and Joachim Buhmann and Jan Puzicha},
  title = {On Learning Texture Edge Detectors},
  booktitle = {International Conference on Image Processing (ICIP'00)},
  year = {2000},
  pages = {877-880}
}
Zöller, T. & Buhmann, J. M. Active Learning for Hierarchical Pairwise Data Clustering 2000 ICPR'00   inproceedings [PDF]  
Abstract: Pairwise data clustering is a well founded grouping technique based on relational data of objects which has a widespread application domain. However, its applicability suffers from the disadvantageous fact that N objects give rise to N(N-1)/2 relations. To cure this unfavorable scaling, techniques to sparsely sample the relations have been developed. Yet a randomly chosen subset of the data might not grasp the structure of the complete data set. To overcome this deficit, we use active learning methods from the field of Statistical Decision Theory. Extending on existing approaches we present a novel algorithm for actively learning hierarchical group structures based on mean field annealing optimization.
BibTeX:
@inproceedings{Zoller2000,
  author = {Thomas Z\"oller and Joachim M. Buhmann},
  title = {Active Learning for Hierarchical Pairwise Data Clustering},
  booktitle = {ICPR'00},
  year = {2000},
  volume = {2},
  pages = {186-189}
}
Buhmann, J. & Held, M. On the optimal number of clusters in histogram clustering 1999 Jahrestagung 99 / Gesellschaft fuer Klassifikation   inproceedings [PDF]  
Abstract: Unsupervised learning algorithms are designed to extract structure from data samples. The quality of a structure is measured by a cost function which is usually minimized to infer optimal parameters characterizing the hidden structure in the data. Reliable and robust inference requires a guarantee that extracted structures are typical for the data source, i.e., similar structures have to be extracted from a second sample set of the same data source. Lack of robustness is known as overfitting from the statistics and the machine learning literature. In this paper the overfitting phenomenon is characterized for a class of histogram clustering models which play a prominent role in information retrieval, linguistic and computer vision applications. Learning algorithms with robustness to sample fluctuations are derived from large deviation theory and the maximum entropy principle for the learning process. The theory validates continuation methods like simulated or deterministic annealing as robust approximation schemes from a statistical learning theory point of view. It predicts the optimal approximation quality, parameterized by a minimal finite temperature, for given covering numbers of the hypothesis class for structures. Monte Carlo simulations are presented to support the underlying inference principle.
BibTeX:
@inproceedings{Buhmann1999a,
  author = {Joachim Buhmann and Marcus Held},
  title = {On the optimal number of clusters in histogram clustering},
  booktitle = {Jahrestagung 99 / Gesellschaft fuer Klassifikation},
  year = {1999}
}
Buhmann, J. M., Malik, J. & Perona, P. Image Recognition: Visual Grouping, Recognition and Learning 1999 National Academy of Science   inproceedings [PDF]  
Abstract: Vision extract useful information from images. Reconstructing the three-dimensional structure of our environment and recognizing the objects that populate it are among the most important functions of our visual system. Computer vision researchers study the computational principles of vision and aim at designing algorithms that reproduce these functions. Vision is difficult: the same scene may give rise to very different images, depending on illumination and viewpoint. Typically, an astronomical number of hypotheses exist that in principle have to be analyzed to infer a correct scene description. Moreover, image information might be extracted at different levels of spatial and logical resolution dependent on the image processing task. Knowledge of the world allows the visual system to limit the amount of ambiguity and to greatly simplify isual computations. We discuss how simple properties of the world are captured by the Gestalt rules of grouping, how the visual system may learn and organize models of objects for recognition, and how one may control the complexity of the description that the visual system computes.
BibTeX:
@inproceedings{Buhmann1999,
  author = {Joachim M. Buhmann and Jitendra Malik and Pietro Perona},
  title = {Image Recognition: Visual Grouping, Recognition and Learning},
  booktitle = {National Academy of Science},
  year = {1999},
  volume = {96}
}
Held, M., Puzicha, J. & Buhmann, J. Visualizing Group Structure 1999 Advances in Neural Information Processing Systems 11 (NIPS'98)   inproceedings [PDF]  
Abstract: Cluster analysis is a fundamental principle in exploratory data analysis, providing the user with a description of the group structure of given data. A key problem in this context is the interpretation and visualization of clustering solutions in high-dimensional or abstract data spaces. In particular, fuzzy or probabilistic descriptions of the group structure, essential to capture inter-cluster relations, are hardly assessable by simple inspection of the probabilistic assignment variables. We present a novel approach for the visualization of probabilistic group structure based on a statistical model of the object assignments which have been observed or estimated by a probabilistic clustering procedure. The objects or data points are embedded in a low dimensional Euclidean space by approximating the observed data statistics with a Gaussian mixture model. The algorithm provides a new approach to the visualization of the inherent structure for a broad variety of data types, e.g. histogram data, proximity data and co-occurrence data. To demonstrate the power of the approach, histograms of textured images are visualized as a large-scale data mining application.
BibTeX:
@inproceedings{Held1999,
  author = {Marcus Held and Jan Puzicha and Joachim Buhmann},
  title = {Visualizing Group Structure},
  booktitle = {Advances in Neural Information Processing Systems 11 (NIPS'98)},
  year = {1999}
}
Hermes, L., Frieauff, D., Puzicha, J. & Buhmann, J. M. Support Vector Machines for Land Usage Classification in Landsat TM Imagery 1999 IEEE International Geoscience and Remote Sensing Symposium   inproceedings [PDF]  
Abstract: Land usage classification is an essential part of many remote sensing applications for mapping, inventory, and yield estimation. In this contribution, we evaluate the potential of the recently introduced support vector machines for remote sensing applications. Moreover, we expand this discriminative technique by a novel Bayesian approach to estimate the confidence of each classification. These estimates are combined with a priori konowledge about topological relations of class labels using a contextual classification step based on the iterative conditional mode algorithm (ICM). As shown for Landsat TM imagery, this strategy is highly competitive and outperforms several commonly used classification schemes.
BibTeX:
@inproceedings{Hermes1999,
  author = {Lothar Hermes and Dieter Frieauff and Jan Puzicha and Joachim M. Buhmann},
  title = {Support Vector Machines for Land Usage Classification in Landsat TM Imagery},
  booktitle = {IEEE International Geoscience and Remote Sensing Symposium},
  year = {1999},
  volume = {1},
  pages = {348-350}
}
Hofmann, T., Puzicha, J. & Jordan, M. Learning from Dyadic Data 1999 Advances in Neural Information Processing Systems 11 (NIPS'98)   inproceedings [PDF]  
Abstract: Dyadic data refers to a domain with two finite sets of objects in which observations are made for dyads, i.e., pairs with one element from either set. This type of data arises naturally in many application ranging from computational linguistics and information retrieval to preference analysis and computer vision. In this paper we present a systematic, domain-independent framework of learning from dyadic data by statistical mixture models. Our approach covers different models with flat and hierarchical latent class structures. We propose an annealed version of the standard EM algorithm for model fitting which is empirically evaluated on a variety of data sets from different domains.
BibTeX:
@inproceedings{Hofmann1999,
  author = {Thomas Hofmann and Jan Puzicha and Michael Jordan},
  title = {Learning from Dyadic Data},
  booktitle = {Advances in Neural Information Processing Systems 11 (NIPS'98)},
  year = {1999},
  pages = {466-472}
}
Klock, H. & Buhmann, J. M. Data Visualization by Multidimensional Scaling: A Deterministic Annealing Approach 1999 Pattern Recognition   article [PDF]  
Abstract: Multidimensional scaling addresses the problem how proximity data can be faithfully visualized as points in a low-dimensional Euclidian space. The quality of a data embedding is measured by a stress function which compares proximity values with Euclidian distances of the respective points. The corresponding minimization problem is non-convex and sensitive to local minima. We present a novel deterministic annealing algorithm for the frequently used objective SSTRESS and for Sammon mapping, derived in the framework of maximum entropy estimation. Experimental results demonstrate the superiority of our optimization technique compared to conventional gradient descent methods.
BibTeX:
@article{Klock1999,
  author = {Hansj\"org Klock and Joachim M. Buhmann},
  title = {Data Visualization by Multidimensional Scaling: A Deterministic Annealing Approach},
  journal = {Pattern Recognition},
  year = {1999},
  volume = {33},
  number = {4},
  pages = {651-669}
}
Puzicha, J. & Buhmann, J. M. Multiscale Annealing for Grouping and Unsupervised texture Segmentation 1999 Computer Vision and Image Understanding (CVIU)   article [PDF]  
Abstract: We derive real-time global optimization methods for several clustering optimization problems commonly used in unsupervised texture segmentation. Speed is achieved by exploiting the image neighborhood relation of features to design a multiscale optimization technique, while accuracy and global optimization properties are gained using annealing techniques. Coarse grained cost functions are derived for central and histogram-based clustering as well as several sparse proximity-based clustering methods.For optimization deterministic annealing algorithms are applied. Annealing schedule, coarse-to-fine optimization and the estimated number of segments are tightly coupled by a statistical convergence criterion derived from computational learning theory. The notion of optimization scale parametrized by a computational temperature is thus unified with the scales defined by the image resolution and the model or segment complexity.The algorithms are benchmarked on Brodatz-like micro-texture mixtures. Results are presented for an autonomous robotics application. Extensions are discussed in the context of prestructuring large image databases valuable for fast and reliable image retrieval.
BibTeX:
@article{Puzicha1999d,
  author = {Jan Puzicha and Joachim M. Buhmann},
  title = {Multiscale Annealing for Grouping and Unsupervised texture Segmentation},
  journal = {Computer Vision and Image Understanding (CVIU)},
  year = {1999},
  volume = {76},
  number = {3},
  pages = {213-230}
}
Puzicha, J., Hofmann, T. & Buhmann, J. M. A Theory of Proximity Based Clustering: Structure Detection by Optimization 1999 Pattern Recognition   article [PDF]  
Abstract: In this paper, a systematic optimization approach for clustering proximity or similarity data is developed. Starting from fundamental invariance and robustness properties, a set of axioms is proposed and discussed to distinguish different cluster compactness and separation criteria. The approach covers the case of sparse proximity matrices, and is extended to nested partitionings for hierarchical data clustering. To solve the associated optimization problems, a rigorous mathematical framework for deterministic annealing and mean-field approximation is presented. Efficient optimization heuristics are derived in a canonical way, which also clarifies the relation to stochastic optimization by Gibbs sampling. Similarity-based clustering techniques have a broad range of possible applications in computer vision, pattern recognition, and data analysis. As a major practical application we present a novel approach to the problem of unsupervised texture segmentation, which relies on statistical tests as a measure of homogeneity. The quality of the algorithms is empirically evaluated on a large collection of Brodatz-like micro-texture Mondrians and on a set of real-word images. To demonstrate the broad usefulness of the theory of proximity based clustering the performances of different criteria and algorithms are compared on an information retrieval task for a document database. The superiority of optimization algorithms for clustering is supported by extensive experiments.
BibTeX:
@article{Puzicha1999,
  author = {Jan Puzicha and Thomas Hofmann and Joachim M. Buhmann},
  title = {A Theory of Proximity Based Clustering: Structure Detection by Optimization},
  journal = {Pattern Recognition},
  year = {1999},
  volume = {33},
  number = {4},
  pages = {617-634}
}
Puzicha, J., Hofmann, T. & Buhmann, J. M. Histogram Clustering for Unsupervised Image Segmentation 1999 IEEE International Conference on Computer Vision and Pattern Recognition (CVPR'99)   inproceedings [PDF]  
Abstract: This paper introduces a novel statistical mixture model for probabilistic grouping of distributional (histogram) data. Adopting the Bayesian framework, we propose to perform annealed maximum a posteriori estimation to compute optimal clustering solutions. In order to accelerate the optimization process, an efficient multiscale formulation is developed. We present a prototypical application of this method for the unsupervised segmentation of textured images based on local distributions of Gabor coefficients. Benchmark results indicate superior performance compared to K-means clustering and proximity-based algorithms.
BibTeX:
@inproceedings{Puzicha1999b,
  author = {Jan Puzicha and Thomas Hofmann and Joachim M. Buhmann},
  title = {Histogram Clustering for Unsupervised Image Segmentation},
  booktitle = {IEEE International Conference on Computer Vision and Pattern Recognition (CVPR'99)},
  year = {1999},
  pages = {602-608}
}
Puzicha, J., Hofmann, T. & Buhmann, J. M. Histogram Clustering for Unsupervised Segmentation and Image Retrieval 1999 Pattern Recognition Letters   article [PDF]  
Abstract: This paper introduces a novel statistical latent class model for probabilistic grouping of distributional and histogram data. Adopting the Bayesian framework, we propose to perform annealed maximum a posteriori estimation to compute optimal clustering solutions. In order to accelerate the optimization process, an efficient multiscale formulation is developed. We present a prototypical application of this method for unsupervised segmentation of textured images based on local distributions of Gabor coefficients. Benchmark results indicate superior performance compared to K-means clustering and proximity-based algorithms. In a second application the histogram clustering method is utilized to structure image databases for improved image retrieval.
BibTeX:
@article{Puzicha1999c,
  author = {Jan Puzicha and Thomas Hofmann and Joachim M. Buhmann},
  title = {Histogram Clustering for Unsupervised Segmentation and Image Retrieval},
  journal = { Pattern Recognition Letters},
  year = {1999},
  volume = {20},
  pages = {899-909}
}
Puzicha, J., Rubner, Y., Tomasi, C. & Buhmann, J. M. Empirical Evaluation of Dissimilarity Measures for Color and Texture 1999 IEEE International Conference on Computer Vision (ICCV'99)   inproceedings [PDF]  
Abstract: This paper empirically compares nine image disimilarity measures that are based on distributions of color and texture features summarizing over 1,000 CPU hours of computational experiments. Ground truth is collected via a novel random sampling scheme for color, and via an image partitioning method for texture. Quantitative performance evaluations are given for classification, image retrieval and, segmentation tasks, and for a wide variety of dissimilarity measures. It is demonstrated how the selection of a measure, based on large scale evaluation, substantially improves the quality of classification, retrieval , and unsupervised segmentation of color and texturte images.
BibTeX:
@inproceedings{Puzicha1999a,
  author = {Jan Puzicha and Yossi Rubner and Carlo Tomasi and Joachim M. Buhmann},
  title = {Empirical Evaluation of Dissimilarity Measures for Color and Texture},
  booktitle = {IEEE International Conference on Computer Vision (ICCV'99)},
  year = {1999},
  pages = {1165-1173}
}
Buhmann, J. M. Empirical Risk Approximation: An Induction Principle for Unsupervised Learning 1998   techreport [PDF]  
Abstract: Unsupervised learning algorithms are designed to extract structure from data without reference to explicit teacher information. The quality of the infered structure is determined by a quality function which guides the search and validation process of structure in data. This paper proposes EMPIRICAL RISK APPROXIMATION as a new induction principle for unsupervised learning. epsilon-covers are used to coarsen the hypothesis class of possible structures. The complexity of the unsupervised learning models are automatically controlled by large deviation bounds. The maximum entropy principle with deterministic annealing as an efficient search strategy arises from the empirical risk approximation principle as the optimal inference strategy for large learning problems. Parameter selection of learnable data structures is demonstrated for the case of K-means clustering.
BibTeX:
@techreport{Buhmann1998a,
  author = {Joachim M. Buhmann},
  title = {Empirical Risk Approximation: An Induction Principle for Unsupervised Learning},
  year = {1998},
  number = {IAI-TR-98-3}
}
Buhmann, J. M., Fellner, D., Held, M., Ketterer, J. & Puzicha, J. Dithered Color Quantization 1998 Computer Craphics Forum   article  
Abstract: Image quantization and digital halftoning are fundamental problems in computer graphics, typically performed when displaying high-color images on non-truecolor devices. Both steps are generally performed sequentially and, in most cases, independent of each other. Color quantization with a pixel-wise defined distortion measure and the dithering process with its local neighborhood typically optimize different quality criteria or, frequently, follow a heuristic without reference to any quality measure. In this paper we propose a new method to simultaneously quantize and dither color images. The method is based on a rigorous cost-function approach which optimizes a quality criterion derived from a generic model of human perception.A highly efficient algorithm for optimization based on a multiscale method is developed for the dithered color quantization cost function. The quality criterion and the optimization algorithms are evaluated on a representative set of artificial and real-world images as well as a collection of icons. A significant image quality improvement is observed compared to standard color reduction approaches.
BibTeX:
@article{Buhmann1998,
  author = {Joachim M. Buhmann and Dieter Fellner and Markus Held and Jens Ketterer and Jan Puzicha},
  title = {Dithered Color Quantization},
  journal = {Computer Craphics Forum},
  year = {1998},
  volume = {17},
  number = {3},
  pages = {219-231}
}
Hofmann, T. & Buhmann, J. M. Competitive Learning Algorithms for Robust Vector Quantization 1998 IEEE Trans. on Signal Processing   article [PDF]  
Abstract: The efficient representation and encoding of signals with limited resources, e.g., finite storage capacity and restricted transmissionbandwidth, is a fundamental problem in technical as well as biological information processing systems. Typically, under realistic circumstances, the encoding and communication of messages has to deal with different sources of noise and disturbances. In this paper, we propose a unifying approach to data compression by robust vector quantization, which explicitly deals with channel noise, bandwidth limitations, and random elimination of prototypes. The resulting algorithm is able to limit the detrimental effect of noise in a very general communication scenario. In addition, the presented model allows us to derive a novel competitive neural networks algorithm, which covers topology preserving feature maps, the so-called neural-gas algorithm, and the maximum entropy soft-max rule as special cases. Furthermore, continuation methods based on these noise models improve the codebook design by reducing the sensitivity to local minima. We show an exemplary application of the novel robust vector quantization algorithm to image compression for a teleconferencing system.
BibTeX:
@article{Hofmann1998,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {Competitive Learning Algorithms for Robust Vector Quantization},
  journal = {IEEE Trans. on Signal Processing},
  year = {1998},
  volume = {46},
  number = {6},
  pages = {1665-1675}
}
Hofmann, T. & Puzicha, J. Mixture Models for Co-occurrence and Histogram Data 1998 International Conference Pattern Recognition   inproceedings [PDF]  
Abstract: Modeling and predicting co-occurrences of events is a fundamental problem of unsupervised learning. In this contribution, we develop a general statistical framework for analyzing co-occurrence data based on probabilistic clustering by mixture models. More specifically, we discuss three models which pursue different modeling goals and which differ in the way they define the probabilistic partitioning of the observations. Adopting the maximum likelihood principle, annealed EM algorithms are derived for parameter estimation. From the class of potential applications in pattern recognition and data analysis, we have chosen document retrieval, language modeling, and unsupervised texture segmentation to test and evaluate the proposed algorithms.
BibTeX:
@inproceedings{Hofmann1998a,
  author = {Thomas Hofmann and Jan Puzicha},
  title = {Mixture Models for Co-occurrence and Histogram Data},
  booktitle = {International Conference Pattern Recognition},
  year = {1998},
  pages = {192-194}
}
Hofmann, T. & Puzicha, J. Unsupervised Learning from Dyadic Data 1998   techreport [PDF]  
Abstract: Dyadic data refers to a domain with two finite sets of objects in which observations are made for dyads, i.e., pairs with one element from either set. This includes event co-occurrences, histogram data, and single stimulus preference data as special cases. Dyadic data arises naturally in many applications ranging from computational linguistics and information retrieval to preference analysis and computer vision. In this paper, we present a systematic, domain-independent framework for unsupervised learning from dyadic data by statistical mixture models. Our approach covers different models with flat and hierarchical latent class structures and unifies probabilistic modeling and structure discovery. Mixture models provide both, a parsimonious yet flexible parameterization of probability distributions with good generalization performance on sparse data, as well as structural information about data-inherent grouping structure. We propose an annealed version of the standard Expectation Maximization algorithm for model fitting which is empirically evaluated on a variety of data sets from different domains.
BibTeX:
@techreport{Hofmann1998b,
  author = {Thomas Hofmann and Jan Puzicha},
  title = {Unsupervised Learning from Dyadic Data},
  year = {1998},
  number = {ICSI-TR-98-042}
}
Hofmann, T. & Puzicha, J. Statistical Models for Co-occurrence Data 1998   techreport  
Abstract: Modeling and predicting co-occurrences of events is a fundamental problem of unsupervised learning. In this contribution we develop a statistical framework for analyzing co-occurrence data in a general setting where elementary observations are joint occurrences of pairs of abstract objects from two finite sets. The main challenge for statistical models in this context is to overcome the inherent data sparseness and to estimate the probabilities for pairs which were rarely observed or even unobserved in a given sample set. Moreover, it is often of considerable interest to extract grouping structure or to find a hierarchical data organization. A novel family of mixture models is proposed which explain the observed data by a finite number of shared aspects or clusters. This provides a common framework for statistical inference and structure discovery and also includes several recently proposed models as special cases. Adopting the maximum likelihood principle, EM algorithms are derived to fit the model parameters. We develop improved versions of EM which largely avoid overfitting problems and overcome the inherent locality of EM-based optimization.Among the broad variety of possible applications, e.g., in information retrieval, natural language processing, data mining, and computer vision, we have chosen document retrieval, the statistical analysis of noun/adjective co-occurrence and the unsupervised segmentation of textured images to test and evaluate the proposed algorithms.
BibTeX:
@techreport{Hofmann1998d,
  author = {Thomas Hofmann and Jan Puzicha},
  title = {Statistical Models for Co-occurrence Data},
  year = {1998},
  number = {1625}
}
Hofmann, T., Puzicha, J. & Buhmann, J. M. Unsupervised Texture Segmentation in a Deterministic Annealing Framework 1998 IEEE Transactions on Pattern Analysis and Machine Intelligence   article [PDF]  
Abstract: We present a novel optimization framework for unsupervised texture segmentation that relies on statistical tests as a measure of homogeneity. Texture segmentation is formulated as a data clustering problem based on sparse proximity data. Dissimilarities of pairs of textured regions are computed from a multi-scale Gabor filter image representation. We discuss and compare a class of clustering objective functions which is systematically derived from invariance principles. As a general optimization framework we propose deterministic annealing based on a mean-field approximation. The canonical way to derive clustering algorithms within this framework as well as an efficient implementation of mean-field annealing and the closely related Gibbs sampler are presented. We apply both annealing variants to Brodatz-like micro-texture mixtures and real-word images.
BibTeX:
@article{Hofmann1998c,
  author = {Thomas Hofmann and Jan Puzicha and Joachim M. Buhmann},
  title = {Unsupervised Texture Segmentation in a Deterministic Annealing Framework},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {1998},
  volume = {20},
  number = {8}
}
Ketterer, J., Puzicha, J., Held, M., Fischer, M., Buhmann, J. M. & Fellner, D. On Spatial Quantization of Color Images 1998 European Conference on Computer Vision   inproceedings [PDF]  
Abstract: Image quantization and dithering are fundamental image processing problems in computer vision and graphics. Both steps are generally performed sequentially and, in most cases, independent of each other. Color quantization with a pixel-wise defined distortion measure and the dithering process with its local neighborhood typically optimize different quality criteria or, frequently, follow a heuristic approach without reference to any quality measure. In this paper we propose a new model to simultaneously quantize and dither color images. The method is based on a rigorous cost-function approach which optimizes a quality criterion derived from a simplified model of human perception. Optimizations are performed by an efficient multiscale procedure which substantially alleviates the computational burden. The quality criterion and the optimization algorithms are evaluated on a representative set of artificial and real-world images thereby showing a significant image quality improvement over standard color reduction approaches.
BibTeX:
@inproceedings{Ketterer1998,
  author = {Jens Ketterer and Jan Puzicha and Marcus Held and Martin Fischer and Joachim M. Buhmann and Dieter Fellner},
  title = {On Spatial Quantization of Color Images},
  booktitle = {European Conference on Computer Vision},
  year = {1998},
  pages = {563-577}
}
Puzicha, J., Hofmann, T. & Buhmann, J. M. Discrete Mixture Models for Unsupervised Image Segmentation 1998 Jahrestagung der Deutsche Arbeitsgemeinschaft für Mustererkennung (DAGM'98)   inproceedings [PDF]  
Abstract: This paper introduces a novel statistical mixture model for probabilistic clustering of histogram data and, more generally, for the analysis of discrete co-occurrence data. Adopting the maximum likelihood framework, an alternating maximization algorithm is derived which is combined with annealing techniques to overcome the inherent locality of alternating optimization schemes. We demonstrate an application of this method to the unsupervised segmentation of textured images based on local empirical distributions of Gabor coefficients. In order to accelerate the optimization process an efficient multiscale formulation is utilized. We present benchmark results on a representative set of Brodatz mondrians and real-world images.
BibTeX:
@inproceedings{Puzicha1998,
  author = {Jan Puzicha and Thomas Hofmann and Joachim M. Buhmann},
  title = {Discrete Mixture Models for Unsupervised Image Segmentation},
  booktitle = {Jahrestagung der Deutsche Arbeitsgemeinschaft f\"ur Mustererkennung (DAGM'98)},
  year = {1998}
}
Buhmann, J. M. Stochastic Algorithms for Exploratory Data Analysis: Data Clustering and Data Visualization 1997   inbook [PDF]  
Abstract: Iterative, EM-type algorithms for data clustering and data visualization are derived on the basis of the maximum entropy principle. These algorithms allow the data analyst to detect structure in vectorial or relational data. Conceptually, the clustering and visualization procedures are formulated as combinatorial or continuous optimization problems which are solved by stochastic optimization.
BibTeX:
@inbook{Buhmann1997a,
  author = {Joachim M. Buhmann},
  title = {Stochastic Algorithms for Exploratory Data Analysis: Data Clustering and Data Visualization},
  year = {1997},
  pages = {333-333}
}
Buhmann, J. M. & Hoffman, T. Robust Vector Quantization by Competitive Learning 1997 International Conference on Acoustics, Speech and Signal Processing ICASSP'97   inproceedings [PDF]  
Abstract: Competitive neural networks can be used to efficiently quantize image and video data. We discuss a novel class of vector quantizers which perform noise robust data compression. The vector quantizers are trained to simultaneously compensate channel noise and code vector elimination noise. The training algorithm to estimate code vectors is derived by the maximum entropy principle in the spirit of deterministic annealing. We demonstrate the performance of noise robust codebooks with compression results for a teleconferencing system on the basis of a wavelet image representation.
BibTeX:
@inproceedings{Buhmann1997,
  author = {Joachim M. Buhmann and Thomas Hoffman},
  title = {Robust Vector Quantization by Competitive Learning},
  booktitle = {International Conference on Acoustics, Speech and Signal Processing ICASSP'97},
  year = {1997}
}
Held, M. & Buhmann, J. M. Unsupervised On-Line Learning of Decision Trees for Hierarchical Data Analysis 1997 Advances in Neural Information Processing Systems (NIPS)   inproceedings [PDF]  
Abstract: An adaptive on-line algorithm is proposed to estimate hierarchical data structures for non-stationary data sources. The approach is based on the principle of minimum cross entropy to derive a decision tree for data clustering and t employs a metalearning idea (learning of learning) to adapt to changes in data characteristics. Its efficiency is demonstrated by grouping non-stationary artifical data and by hierarchical segmentation of LANDSAT images.
BibTeX:
@inproceedings{Held1997,
  author = {Marcus Held and Joachim M. Buhmann},
  title = {Unsupervised On-Line Learning of Decision Trees for Hierarchical Data Analysis},
  booktitle = {Advances in Neural Information Processing Systems (NIPS)},
  year = {1997},
  pages = {514-520}
}
Hofmann, T. & Buhmann, J. M. Pairwise Data Clustering by Deterministic Annealing 1997 IEEE Transactions on Pattern Analysis and Machine Intelligence   article [PDF]  
Abstract: Partitioning a data set and extracting hidden structure from the data arises in different application areas of pattern recognition and image processing. Pairwise data clustering is a combinatorial optimisation method for data grouping which extracts structure from proximity data. We descibe a deterministic annealing approach to pairwise data clustering which shares the robustness properties of maximum entropy inference. The resulting Gibbs probability distributions are estimated by meanfield approximation, a well-known technique from statistical physics. A new algorithm to group dissimilarity data and to simultanously embed these data in a Euclidian vector space is discussed which can be used for dimension reduction and data visualisation. The suggested algorithms have been implemented to analyse dissimilarity data from protein analysis and from linguistics. Furthermore, pairwise data clustering is used to segment textured images.
BibTeX:
@article{Hofmann1997a,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {Pairwise Data Clustering by Deterministic Annealing},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {1997},
  volume = {19},
  number = {1},
  pages = {1-14}
}
Hofmann, T. & Buhmann, J. M. Active Data Clustering 1997 Advances in Neural Information Processing Systems (NIPS)   inproceedings [PDF]  
Abstract: Active Data Clustering is a novel technique for clustering proximity data which utilizes principles from sequential experiment design in order to interleave data generation and data analysis. The proposed active data sampling strategy is based on the expected value of information, a concept rooting in statistical decision theory. This is considered to be an important step towards the analysis of large-scale data sets, because it offers a way to overcome the inherent data sparseness of proximity data. We present applications to unsupervised texture segmentation in computer vision and information retrieval in document databases.
BibTeX:
@inproceedings{Hofmann1997b,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {Active Data Clustering},
  booktitle = {Advances in Neural Information Processing Systems (NIPS)},
  year = {1997},
  pages = {528-534}
}
Hofmann, T., Puzicha, J. & Buhmann, J. M. Deterministic Annealing for Unsupervised Texture Segmentation 1997 International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR'97)   inproceedings [PDF]  
Abstract: In this paper a rigorous mathematical framework of deterministic annealing and mean-field approximation is presented for a general class of partitioning, clustering and segmentation problems. We describe the canonical way to derive efficient optimization heuristics, which have a broad range of possible applications in computer vision, pattern recognition and data analysis. In addition, we prove novel convergence results. As a major practical application we present a new approach to the problem of unsupervised texture segmentation which relies on statistical tests as a measure of homogeneity. More specifically, this results in a formulation of texture segmentation as a pairwise data clustering problem with a sparse neighborhood structure. We discuss and compare different clustering objective functions, which are systematically derived from invariance principles. The quality of the novel algorithms is empirically evaluated on a large database of Brodatz-like micro-texture mixtures and on a representative set of real-word images.
BibTeX:
@inproceedings{Hofmann1997,
  author = {Thomas Hofmann and Jan Puzicha and Joachim M. Buhmann},
  title = {Deterministic Annealing for Unsupervised Texture Segmentation},
  booktitle = {International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR'97)},
  year = {1997},
  pages = {213-228}
}
Hofmann, T., Puzicha, J. & Buhmann, J. M. An Optimization Approach to Unsupervised Hierarchical Texture Segmentation 1997 International Conference on Image Processing (ICIP '97)   inproceedings [PDF]  
Abstract: In this paper we introduce a novel optimization framework for hierarchical data clustering which is applied to the problem of unsupervised texture segmentation. The proposed objective function assesses the quality of an image partitioning simultaneously at different resolution levels. A novel model selection criterion to select significant image structures from various scales is proposed. As an efficient deterministic optimization heuristic a mean-field annealing algorithm is derived.
BibTeX:
@inproceedings{Hofmann1997c,
  author = {Thomas Hofmann and Jan Puzicha and Joachim M. Buhmann},
  title = {An Optimization Approach to Unsupervised Hierarchical Texture Segmentation},
  booktitle = {International Conference on Image Processing (ICIP '97)},
  year = {1997}
}
Klock, H. & Buhmann, J. M. Multidimensional Scaling by Deterministic Annealing 1997 International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR'97)   inproceedings [PDF]  
Abstract: Multidimensional scaling addresses the problem how proximity data can be faithfully visualized as points in a low-dimensional Euclidian space. The quality of a data embedding is measured by a cost function called stress which compares proximity values with Euclidian distances of the respective points. We present a novel deterministic annealing algorithm to efficiently determine embedding coordinates for this continuous optimization problem. Experimental results demonstrate the superiority of the optimization technique compared to conventional gradient descent methods. Furthermore, we propose a transformation of dissimilarities to reduce the mismatch between a high-dimensional data space and a low-dimensional embedding space.
BibTeX:
@inproceedings{Klock1997,
  author = {Hansj\"org Klock and Joachim M. Buhmann},
  title = {Multidimensional Scaling by Deterministic Annealing},
  booktitle = {International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR'97)},
  year = {1997},
  pages = {245-260}
}
Klock, H., Polzer, A. & Buhmann, J. M. Region-Based Motion Compensated 3D-Wavelet Transform Coding of Video 1997 IEEE International Conference on Image Processing (ICIP '97)   inproceedings [PDF]  
Abstract: We present a low bit-rate video compression system that integrates region-based coding with a spatio-temporal wavelet transform. The proposed system is designed for monitoring and video-phone applications. It distinguishes between moving foreground and static background, but image segmentation might also be based on other sources. The regions are encoded in separate layers using a chroma-keying technique that allows a controlled lossy recovery of the boundaries. A 3D wavelet transform is applied to a group of frames of the predictor residual signal. Statistical dependencies of transform coefficients extracted from different image subbands are captured by conditional probability models. Without the layered coding, the system performs superior compared to the recent H.263 standard for very low bit-rate coding. The layered coding causes a small degradation in visual quality at the same bit-rate.
BibTeX:
@inproceedings{Klock1997a,
  author = {Hans-J\"org Klock and Andreas Polzer and Joachim M. Buhmann},
  title = {Region-Based Motion Compensated 3D-Wavelet Transform Coding of Video},
  booktitle = {IEEE International Conference on Image Processing (ICIP '97)},
  year = {1997}
}
Polzer, A., Klock, H. & Buhmann, J. Video-Coding by Region-Based Motion Compensation and Spatio-temporal wavelet transform 1997 IEEE International Conference on Image Processing (ICIP '97)   inproceedings [PDF]  
Abstract: We present a low bit-rate video compression system that integrates region-based motion estimation and motion compensation with spatio-temporal wavelet transform coding. The paper focusses on the motion segmentation module, which determines maximum likelihood estimates of region motion parameters employing the Expectation Maximization (EM) algorithm and mean field techniques. Preliminary results show the competitiveness of this approach.
BibTeX:
@inproceedings{Polzer1997,
  author = {Andreas Polzer and Hans-J\"org Klock and Joachim Buhmann},
  title = {Video-Coding by Region-Based Motion Compensation and Spatio-temporal wavelet transform},
  booktitle = {IEEE International Conference on Image Processing (ICIP '97)},
  year = {1997}
}
Puzicha, J. & Buhmann, J. M. Multiscale Annealing for Real-Time Unsupervised Texture Segmentation 1997   techreport [PDF]  
Abstract: We derive real-time global optimization methods for several clustering optimization problems used in unsupervised texture segmentation. Speed is achieved by exploiting the topological relation of features to design a multiscale optimization technique, while accuracy and global optimization properties are gained using a deterministic annealing method. Coarse grained cost functions are derived for both central and sparse pairwise clustering, where the problem of coarsening sparse random graphs is solved by the concept of structured randomization. Annealing schedules and coarse-to-fine optimization are tightly coupled by a statistical convergence criterion derived from computational learning theory. The algorithms are benchmarked on Brodatz-like micro-texture mixtures. Results are presented for an autonomous robotics application.
BibTeX:
@techreport{Puzicha1997b,
  author = {Jan Puzicha and Joachim M. Buhmann},
  title = {Multiscale Annealing for Real-Time Unsupervised Texture Segmentation},
  year = {1997},
  number = {IAI-TR-97-4}
}
Puzicha, J., Hofmann, T. & Buhmann, J. M. Deterministic Annealing: Fast Physical Heuristics for Real-Time Optimization of Large Systems 1997 15th IMACS World Conference on Scientific Computation, Modelling and Applied Mathematics   inproceedings [PDF]  
Abstract: This paper systematically investigates the heuristical optimization technique known as deterministic annealing. This method is applicable to a large class of assignment and partitioning problems. Moreover, the established theoretical results, as well as the general algorithmic solution scheme, are largely independent of the objective functions under consideration. Deterministic annealing is derived from strict minimization principles, including a rigorous convergence analysis. We stress the close relation to homotopy methods, and discuss some of the most important strengths and weaknesses in this framework. Optimization results for unsupervised texture segmentation are presented for an autonomous robotics application.
BibTeX:
@inproceedings{Puzicha1997,
  author = {Jan Puzicha and Thomas Hofmann and Joachim M. Buhmann},
  title = {Deterministic Annealing: Fast Physical Heuristics for Real-Time Optimization of Large Systems},
  booktitle = {15th IMACS World Conference on Scientific Computation, Modelling and Applied Mathematics},
  year = {1997}
}
Puzicha, J., Hofmann, T. & Buhmann, J. M. Non-parametric Similarity Measures for Unsupervised Texture Segmentation and Image Retrieval 1997 IEEE International Conference on Computer Vision and Pattern Recognition   inproceedings [PDF]  
Abstract: In this paper we propose and examine non-parametric statistical tests to define similarity and homogeneity measures for textures. The statistical tests are applied to the coefficients of images filtered by a multi-scale Gabor filter bank. We will demonstrate that these similarity measures are useful for both, texture based image retrieval and for unsupervised texture segmentation, and hence offer an unified approach to these closely related tasks. We present results on Brodatz-like micro-textures and a collection of real-word images.
BibTeX:
@inproceedings{Puzicha1997a,
  author = {Jan Puzicha and Thomas Hofmann and Joachim M. Buhmann},
  title = {Non-parametric Similarity Measures for Unsupervised Texture Segmentation and Image Retrieval},
  booktitle = {IEEE International Conference on Computer Vision and Pattern Recognition},
  year = {1997},
  pages = {267-272}
}
Fröhlinghaus, T. & Buhmann, J. M. Real-Time Phase-Based Stereo for a Mobile Robot 1996 First Euromicro Workshop on Advanced Mobile robots   inproceedings [PDF]  
Abstract: The performance of a mobile robot crucially depends on the accuracy, duration and reliability of its sensor interpretation. A major source of information are CCD-cameras which provide a detailed view of the robot's environment. This paper presents a real-time stereo algorithm implemented on the mobile robot Rhino of the University of Bonn. The algorithm exploit the phases of wavelet-filtered image pairs to localize edges and to estimate their disparities with subpixel accuracy. The disparities are computed by an initial search for corresponding points within a given interval and a subsequent measurement of phase-differences. The real-time constraints of autonomous object detection and navigation are fulfilled by partially implementing the stereo algorithm on a pipeline computer Datacube. Experimental results on real world scenes under real world conditions demonstrate the stereo algorithm's robustness and suitability for autonomous robot applications.
BibTeX:
@inproceedings{Frohlinghaus1996,
  author = {Thorsten Fr\"ohlinghaus and Joachim M. Buhmann},
  title = {Real-Time Phase-Based Stereo for a Mobile Robot},
  booktitle = {First Euromicro Workshop on Advanced Mobile robots},
  year = {1996},
  pages = {178-185}
}
Hofmann, T. & Buhmann, J. M. An Annealed Neural Gas Network for Robust Vector Quantization 1996 International Conference on Artificial Neural Networks ICANN96   inproceedings [PDF]  
Abstract: Vector quantization, a central topic in data compression, deals with the problem of encoding an information source or a sample of data vectors by means of a finite codebook, such that the average distortion is minimized. We introduce a common framework, based on maximum entropy inference to derive a deterministic annealing algorithm for robust vector quantization. The objective function for codebook design is extended to take channel noise and bandwidth limitations into account. Formulated as an on--line problem it is possible to derive learning rules for competitive neural networks. The resulting update rule is a generalization of the `neural gas' model. The foundation in coding theory allows us to specify an optimality criterion for the `neural gas' update rule.
BibTeX:
@inproceedings{Hofmann1996,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {An Annealed Neural Gas Network for Robust Vector Quantization},
  booktitle = {International Conference on Artificial Neural Networks ICANN96},
  year = {1996},
  pages = {151-156}
}
Hofmann, T. & Buhmann, J. M. Inferring Hierarchical Clustering Structures by Deterministic Annealing 1996 Second International Conference on Knowledge Discovery and Data Mining KDD'96   inproceedings [PDF]  
Abstract: The unsupervised detection of hierarchical structures is a major topic in unsupervised learning and one of the key questions in data analysis and representation. We propose a novel algorithm for the problem of learning decision trees for data clustering and related problems. In contrast to many other methods based on successive tree growing and pruning, we propose a completely non--greedy technique based on an explicit objective function. Applying the principles of maximum entropy and minimum cross entropy, a deterministic annealing algorithm is derived in a meanfield approximation. This technique allows us to canonically superimpose tree structures and to fit parameters to averaged or `fuzzified' trees.
BibTeX:
@inproceedings{Hofmann1996a,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {Inferring Hierarchical Clustering Structures by Deterministic Annealing},
  booktitle = {Second International Conference on Knowledge Discovery and Data Mining KDD'96},
  year = {1996},
  pages = {363-366}
}
Hofmann, T., Puzicha, J. & Buhmann, J. M. Unsupervised Segmentation of Textured Images by Pairwise Data Clustering 1996 International Conference on Image Processing ( ICIP '96 )   inproceedings [PDF]  
Abstract: We propose a novel approach to unsupervised texture segmentation, which is formulated as a combinatorial optimization problem known as pairwise data clustering with a sparse neighborhood structure. Pairwise dissimilarities between texture blocks are measured in terms of distribution differences of multi--resolution features. The feature vectors are based an a Gabor wavelet image representation.To efficiently solve the data clustering problem a deterministic annealing algorithm based on a meanfield approximation is derived. An application to Brodatz--like microtexture mixtures is shown. We statistically adress the questions of adequacy of the proposed cost function and the quality of the deterministic annealing algorithm compared with its stochastic variants.
BibTeX:
@inproceedings{Hofmann1996b,
  author = {Thomas Hofmann and Jan Puzicha and Joachim M. Buhmann},
  title = {Unsupervised Segmentation of Textured Images by Pairwise Data Clustering},
  booktitle = {International Conference on Image Processing ( ICIP '96 )},
  year = {1996},
  volume = {III},
  pages = {137-140}
}
Klock, H. & Buhmann, J. M. Data Visualization by Multidimensional Scaling: A Deterministic Annealing Approach 1996   techreport  
Abstract: Multidimensional scaling addresses the problem how proximity data can be faithfully visualized as points in a low-dimensional Euclidian space. The quality of a data embedding is measured by a stress function which compares proximity values with Euclidian distances of the respective points. The corresponding minimization problem is non-convex and sensitive to local minima. We present a novel deterministic annealing algorithm for the frequently used objective SSTRESS and for Sammon mapping , derived in the framework of maximum entropy estimation. Experimental results demonstrate the superiority of our optimization technique compared to conventional gradient descent methods.
BibTeX:
@techreport{Klock1996,
  author = {Hansj\"org Klock and Joachim M. Buhmann},
  title = {Data Visualization by Multidimensional Scaling: A Deterministic Annealing Approach},
  year = {1996},
  number = {IAI-TR-96-8}
}
Puzicha, J., Goerke, N. & Eckmiller, R. Optimal Trajectory Generation for an Industrial Robot by Markovian Networks 1996 International Conference on Neural Information Processing (ICONIP'96)   inproceedings [PDF]  
Abstract: Optimal trajectory generation can be represented as a constrained variational problem posed simultaneously in cartesian and joint coordinate spaces. Markovian Networks are presented as a class of neural networks especially suited for solving general constrained variational problems. By incorporation of the kinematic map an iterative network dynamic is created, which gradually converges to an optimal solution. The capability of this approach is demonstrated by generating optimal 6--DOF movements along surfaces for an industrial robot arm with remarkable improvements compared to a standard algorithm.
BibTeX:
@inproceedings{Puzicha1996,
  author = {Jan Puzicha and Nils Goerke and Rolf Eckmiller},
  title = {Optimal Trajectory Generation for an Industrial Robot by Markovian Networks},
  booktitle = {International Conference on Neural Information Processing (ICONIP'96)},
  year = {1996},
  pages = {995-998}
}
Puzicha, J., Hofmann, T. & Buhmann, J. Unsupervised Texture Segmentation on the Basis of Scale Space Features 1996   techreport [PDF]  
Abstract: A novel approach to unsupervised texture segmentation is presented, which is formulated as a combinatorial optimization problem known as sparse pairwise data clustering. Pairwise dissimilarities between texture blocks are measured by scale space features, i.e., multi-resolution edges. These scale space features are computed by a Gabor filter bank tuned to spatial frequencies. To solve the data clustering problem a deterministic annealing technique is applied. This approach is examined from the viewpoint of scale space theory. Based on a meanfield approximation an efficient algorithm is derived. We present an application of the proposed algorithm to Brodatz--like microtexture collages.
BibTeX:
@techreport{Puzicha1996a,
  author = {Jan Puzicha and Thomas Hofmann and Joachim Buhmann},
  title = {Unsupervised Texture Segmentation on the Basis of Scale Space Features},
  year = {1996},
  number = {DIKU 96/19}
}
Buhmann, J. M. Oscillatory Associative Memories 1995 Handbook of Brain Theory & Neural Networks   inbook [PDF]  
Abstract: Assemblies of cooperating binary units have been postulated as the basic building blocks of associative memories. Artificial neural networks with a simple dynamics evolve from initial states to predefined fixed points, the memory traces. Storage and retrieval in these neural networks with fixed point attractors are limited to one attern at a time. Biological and functional considerations especially in olfactory cortex motivate oscillatory neural network dynamics which model experimentally observed neural activity patterns in associative recall tasks. Binding features to stored patterns when several memory traces are coactivated requires a network dynamics with oscillatory or even chaotic activity patterns.
BibTeX:
@inbook{Buhmann1995,
  author = {Joachim M. Buhmann},
  title = {Oscillatory Associative Memories},
  booktitle = {Handbook of Brain Theory \& Neural Networks},
  year = {1995},
  pages = {333-333}
}
Buhmann, J. M., Burgard, W., Cremers, A. B., Fox, D., Hofmann, T., Schneider, F., Strikos, J. & Thrun, S. The Mobile Robot Rhino 1995 AI Magazin   article [PDF]  
Abstract: Rhino was the University of Bonn's entry in the 1994 AAAI mobile robot competition. Rhino is a mobile robot designed for indoor navigation and manipulation tasks. The general scientific goal of the Rhino project is the development and the analysis of autonomous and complex learning systems. This paper briefly describes the major components of the Rhino control software, as they were exhibited at the competition. It also sketches the basic philosophy of the Rhino architecture, and discusses some of the lessons that we learned during the competition.
BibTeX:
@article{Buhmann1995a,
  author = {Joachim M. Buhmann and Wolfram Burgard and Armin B. Cremers and Dieter Fox and Thomas Hofmann and Frank Schneider and Jiannis Strikos and Sebastian Thrun},
  title = {The Mobile Robot Rhino},
  journal = {AI Magazin},
  year = {1995},
  volume = {16},
  number = {1}
}
Hofmann, T. & Buhmann, J. M. Hierarchical Pairwise Data Clustering by Mean-Field Annealing 1995 International Conference on Artificial Neural Networks (ICANN'95)   inproceedings [PDF]  
Abstract: Partitioning a data set and extracting hidden structure arises in different application areas of pattern recognition, data analysis and image processing. We formulate data clustering for data characterized by pairwise dissimilarity values as an assignement problem with an objective function to be minimized. An extension to tree-structured clustering is proposed which allows a hierarchical grouping of data. Deterministic annealing algorithms are derived for unconstrained and tree-structured pairwise clustering.
BibTeX:
@inproceedings{Hofmann1995,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {Hierarchical Pairwise Data Clustering by Mean-Field Annealing},
  booktitle = {International Conference on Artificial Neural Networks (ICANN'95)},
  year = {1995},
  pages = {197-202}
}
Hofmann, T. & Buhmann, J. M. Multidimensional Scaling and Data Clustering 1995 Advances in Neural Information Processing Systems 7 (NIPS'94)   inproceedings [PDF]  
Abstract: Visualizing and structuring pairwise dissimilarity data are difficult combinatorial problems known as multidimensional scaling or pairwise data clustering. Algorithms for embedding dissimilarity data sets in Euclidian space, for clustering these data and for actively selecting data to support the clustering process are discussed in the maximum entropy framework. Active data selection provides a strategy to discover structure in a data set efficiently with partially unknown data.
BibTeX:
@inproceedings{Hofmann1995a,
  author = {Thomas Hofmann and Joachim M. Buhmann},
  title = {Multidimensional Scaling and Data Clustering},
  booktitle = { Advances in Neural Information Processing Systems 7 (NIPS'94)},
  year = {1995},
  pages = {104-111}
}
Buhmann, J. M. & Hofmann, T. A Maximum Entropy Approach to Pairwise Data Clustering 1994 International Conference on Pattern Recognition   inproceedings [PDF]  
Abstract: Partitioning a set of data points which are characterized by their mutual dissimilarities instead of an explicit coordinate representation is a difficult, NP-hard combinatorial optimization problem. We formulate this optimization problem of a pairwise clustering cost function in the maximum entropy framework using a variational principle to derive corresponding data partitionings in a d-dimensional Euclidian space. This approximation solves the embedding problem and the grouping of these data into clusters simultaneously and in a selfconsistent fashion.
BibTeX:
@inproceedings{Buhmann1994,
  author = {Joachim M. Buhmann and Thomas Hofmann},
  title = {A Maximum Entropy Approach to Pairwise Data Clustering},
  booktitle = {International Conference on Pattern Recognition},
  year = {1994},
  pages = {207-212}
}
Buhmann, J. M., Lades, M. & Eeckmann, F. Illumination-Invariant Face Recognition with a Contrast Sensitive Silicon Retina 1994 Advances in Neural Information Processing Systems (NIPS)   inproceedings [PDF]  
Abstract: Changes in lighting conditions strongly effect the performance and reliability of computer vision systems. We report face recognition results under drastically changing lightning conditions for a computer vision system which concurrently uses a contrast sensitive silicon retina and a conventional, gain controlled CCD camera. For both input devices the face recognition system employs an elastic matching algorithm with wavelet based features to classify unknown faces. To assess the effect of analog on-chip preprocessing by the silicon retina the CCD images have been igitally preprocessedwith a bandpass filter to adjust the power spectrum. The silicon retina with its ability to adjust sensitivity increases the recognition rate up to 50 percent. These comparative experiments demonstrate that preprocessing with an analog VLSI silicon retina generates image data enriched with object-constant features.
BibTeX:
@inproceedings{Buhmann1994a,
  author = {Joachim M. Buhmann and Martin Lades and Frank Eeckmann},
  title = {Illumination-Invariant Face Recognition with a Contrast Sensitive Silicon Retina},
  booktitle = {Advances in Neural Information Processing Systems (NIPS)},
  year = {1994},
  volume = {6},
  pages = {769-776}
}
Buhmann, J. M. & Kühnel, H. Vector Quantization with Complexity Costs 1993 IEEE Transactions on Information Theory   article [PDF]  
Abstract: Vector quantization is a data compression method where a set of data points is encoded by a reduced set of reference vectors, the codebook. We discuss a vector quantization strategy which jointly optimizes distortion errors and the codebook complexity, thereby, determining the size of the codebook. A maximum entropy estimation of the cost function yields an optimal number of reference vectors, their positions and their assignement probabilities. The dependence of the codebook density on the data density for different complexity measures is investigated in the limit of asymptotic quantization levels. How different complexity measures influence the efficiency of vector quantization is studied for the task of image compression, i.e., we quantize the wavelet coefficients of gray level images and measure the reconstruction error. Our approach establishes a unifying framework for different quantization methods like K-means clustering and its fuzzy version, entropy constrained vector quantization or topological feature maps and competitive neural networks.
BibTeX:
@article{Buhmann1993a,
  author = {Joachim M. Buhmann and Hans K\"uhnel},
  title = {Vector Quantization with Complexity Costs},
  journal = {IEEE Transactions on Information Theory},
  year = {1993},
  volume = {39},
  pages = {1133-1145}
}
Buhmann, J. M. & Kühnel, H. Complexity Optimized Data Clustering by Competitive Neural Networks 1993 Neural Computation   article  
BibTeX:
@article{Buhmann1993,
  author = {Joachim M. Buhmann and Hans K{\"u}hnel},
  title = {Complexity Optimized Data Clustering by Competitive Neural Networks},
  journal = {Neural Computation},
  year = {1993},
  volume = {5},
  pages = {75-88}
}
Lades, M., Vorbrüggen, J. C., Buhmann, J. M., Lange, J., von der Malsburg, C., Würtz, R. P. & Konen, W. Distortion Invariant Object Recognition in the Dynamic Link Architecture 1993 IEEE Transactions on Computers   article  
BibTeX:
@article{Lades1993,
  author = {Martin Lades and Jan C. Vorbr\"uggen and Joachim M. Buhmann and J\"org Lange and Christoph {von der Malsburg} and Rolf P. W\"urtz and Wolfgang Konen},
  title = {Distortion Invariant Object Recognition in the Dynamic Link Architecture},
  journal = {IEEE Transactions on Computers},
  year = {1993},
  volume = {42},
  pages = {300-311}
}
Arbib, M. A. & Buhmann, J. M. Neural Networks 1992 Encyclopedia of Artificial Intelligence   inproceedings  
BibTeX:
@inproceedings{Arbib1992,
  author = {Michael A. Arbib and Joachim M. Buhmann},
  title = {Neural Networks},
  booktitle = {Encyclopedia of Artificial Intelligence},
  publisher = {John Wiley},
  year = {1992},
  volume = {2},
  pages = {1016-1060}
}
Buhmann, J. M. Neuronale Netzwerke als assoziative Speicher und als Systeme zur Mustererkennung 1988 School: Physik-Department, Technische Universität München   phdthesis  
BibTeX:
@phdthesis{Buhmann1988,
  author = {Joachim M. Buhmann},
  title = {Neuronale Netzwerke als assoziative Speicher und als Systeme zur Mustererkennung},
  school = {Physik-Department, Technische Universit\"at M\"unchen},
  year = {1988}
}

Created by JabRef on 03/12/2008.