Readspeaker menu

Dr. Michael Kemmler

Former Research Associate
kemmler_photo3.png
e-mail: kemmler (dot) michael (at) gmail (dot) com

Studium

  • 2001-2009 student of computer science at FSU Jena
  • specialization: pattern recognition and artificial intelligence
  • student research project: Modelling Similarity Assesments for Gene Expression Cluster Analysis (in German, Modellierung von Ähnlichkeitsbewertungen für die Clusteranalyse von Genexpressionsdaten)
  • diploma thesis: Using Global Image Features as Context Information for Scene and Object Recognition (in German, Verwendung globaler Bildmerkmale als Kontextinformation für Szenen- und Objekterkennung)

Arbeit am Lehrstuhl

  • Assistent for organization and realization of the conference DAGM2009.             
dagm_banner.png
  • Participation in the MikroPlex-Teilprojekt "Anpassung der Bildauswertungsmethoden auf die Bakterien-Lokalisierung" subproject "Adapting Image Analysis Methods for Bacteria Localization" in cooperation with the IPHT and the Institut für Physikalische Chemie.
 mikroplex_banner.jpg












Forschungsthemen/Publikationen


  • One-class classification and novelty detection       

  GPR_mean.jpgGPR_variance.jpg
 In many important task, training examples are often available for only one class. Learning the support of the category under focus is difficult, since training examples from "outside" are not available at all. This problem is known as one-class classification (OCC), novelty detection and outlier detection, to name just a few. The current paper addresses the use of Gaussian Processes for deriving a set of OCC scores. Empirical comparison with Support Vector Data Description (SVDD) on visual object recognition tasks using image kernels indicate the suitability of our approach.
dl2.png Code for one-class classification based on Gaussian process regression (GitHub)
 
  novelbac.jpg
 When classifying microorganisms, the typical closed-world assumption of pattern recognition does not hold. While today's databases are getting more and more complex, an exhaustive training database covering all possible species is not realistic due to the large biodiversity of microbes. In practice, it is therefore of utmost importance to detect novel microbes, which are not present during training. Our work analyzes the use of Gaussian process based one-class classification for this task. For a bacteria dataset comprising 50 different strains, we show its suitability in comparison to other state-of-the-art methods such as Gaussian mixture models, Parzen density estimation, and Support Vector Data Description.  

  rope_example1.jpg
rope_example2.jpg
 Controlling ropes is an important task for ensuring the safety of elevators, bridges, or cable car transportation systems (see the page of Esther Wacker for a larger list of publications related to this topic). In practice, this is regularly done by a human observer who has to inspect every inch of a given rope searching for suspicious anomalies. An automatic procedure for solving this tedious task or for narrowing down the suspicious regions to a small fraction would be beneficial. In this approach, a recently proposed novelty detection method using Gaussian processes in combination with appearance-based HOG features is employed to tackle this problem. Experiments on a challenging ropeway dataset demonstrate that this approach scales favorably with respect to existing state-of-the-art techniques such as Gaussian Mixture Models and Support Vector Data Description.  
  knfst.png
  In current applications of novelty detection, usually more than one class is already given in the training database. While one-class classification methods can be adapted for this setting, e.g., by building multiple one-class models or a single super-class, useful information can be easily lost in this process. This work tackles the multi-class novelty detection problem using the kernel null Foley-Sammon transform (KNFST) by jointly projecting the data into the subspace with maximum inter-class variance (preserving the presumed differences between classes) and zero intra-class variance (facilitating a compact representation of classes). Extensive experiments on Caltech-256 and ImageNet showed that our KNFST novelty detection scheme significantly outperformed several well-known state-of-the-art novelty detection approaches (OVA-SVM, OC-SVDD, OC-GP). Our experiments hence clearly validate the benefits of multi-class novelty detection methods that aim to jointly learn a model for all classes simultaneously.  
  • Classification of Raman spectra        

  ramanclass.jpg
 In this work we are dealing with the classification of microorganisms based on Raman spectroscopy, where the Gaussian Process classifier is investigated for the accurate discrimination of different bacterial species. Moreover, this work also discusses the incorporation of invariances into the Gaussian Process framework and its benefit for small training samples.
 
  • Feature relevance in Raman spectroscopy       

  feature_relevance.jpg
 This paper tackles the problem of finding important features (in the input space) for Raman spectroscopy. Several heuristics derived from supervised classification techniques are employed to generate relevance scores which highlight the importance of features with respect to the underlying classification task.
 
  • Detection of microorganisms in microscopy images       

  microbe_detection_scheme.jpg
 The detection of microorganisms is an essential step in many challenging tasks such as cell recognition and cell tracking. It is often resolved by using fluorescence techniques which highlight biological material. This paper deals with automatically detecting microorganisms using microscopy images where no additional staining has been performed. Level set segmentation and segment-based methods were compared on a brightfield microscopy dataset containing microbe images that are partly contaminated with sand particles. The analysis highlights the benefits and pitfalls of all algorithms and should serve as a basis for more complex settings.
 
  • Efficient Gaussian process classification      

  rdfgp_notext.jpg
 
 This work deals with combining Gaussian process classifiers and deterministic decision trees and random forests for fast and accurate categorization of indoor scenes and large-scale pattern recognition problems. The trade-off between the two contrary goals, accuracy and speed, can be easily integrated into this framework by tuning the maximum number of samples reaching the leaf nodes (which are directly fed into the final GP classifiers). The suitability of this approach was additionally demonstrated for a challenging facade recognition task.
 
  • Facial action unit intensity prediction   

 
  The estimation of facial action units (AUs) is a central step in expression analysis. In this context, usually only presence or absence of AUs are estimated. This paper goes beyond this binary prediction step by efficiently estimating the whole range of available AU activation intensities, allowing for more complex applications (e.g. supporting facial palsy diagnosis). Using a low-dimensional representation derived from active appearance models, extensive experiments on three well-known databases were conducted. While computational experiments showed the efficiency of our approach, we also concluded that regression methods such as Gaussian process regression are generally to be preferred to classification-based prediction methods (nearest neighbor method, SVM).
 
  • Place recognition using a ToF camera   

  place_corridor.jpgplace_studlab.jpg
 We present an approach to place recognition based on an uncalibrated combination of range data of a time-of-flight (ToF) camera and images obtained from a visual sensor. Our system is able to classify the environment in predefined places (e.g. kitchen, corridor, office) by representing the sensor data with various global features.
 

All publications in chronological order