Norbert Krüger

Face recognition by elastic bunch graph matching

We present a system for recognizing human faces from single images out of a large database containing one image per person. Faces are represented by labeled graphs, based on a Gabor wavelet transform. Image graphs of new faces are extracted by an elastic graph matching process and can be compared by a simple similarity function. The system differs from Lades et al. (1993) in three respects. Phase information is used for accurate node positioning. Object-adapted graphs are used to handle large rotations in depth. Image graph extraction is based on a novel data structure, the bunch graph, which is constructed from a small set of sample image graphs.

Collinearity and Parallelism are Statistically Significant Second-Order Relations of Complex Cell Responses

By investigating the second-order statistics of Gabor wavelet responses derived from natural images, we show that collinearity and parallelism are conspicuous relations. We give a precise mathematical characterization of these Gestalt principles by the conditional probability of two responses. Essential for our investigations is a non-linear transformation, initially utilized within the object recognition system [5], which transforms continuous Gabor wavelet responses into a binary code indicating the presence or absence of local oriented line segments.

Improving object recognition by transforming Gabor filter responses

Previous work described a biologically motivated object recognition system with Gabor wavelets as basic feature type. These features are robust against slight distortion, rotation and variation in illumination. We here describe extensions of the system that address image variance due to arbitrary in-plane rotation, substantial scale changes and moderate depth rotation of objects, and to background variation, using simple linear transformation of the Gabor filter responses. The performance of the system is enhanced significantly.

Face Recognition by Elastic Bunch Graph Matching

We present a system for recognizing human faces from single images out of a large database with one image per person. The task is difficult because of image variation in terms of position, size, expression, and pose. The system collapses most of this variance by extracting concise face descriptions in the form of image graphs. In these, fiducial points on the face (eyes, mouth etc.) are described by sets of wavelet components (jets). Image graph extraction is based on a novel approach, the bunch graph, which is constructed from a small set of sample image graphs. Recognition is based on a straight-forward comparison of image graphs. We report recognition experiments on the FERET database and the Bochum database, including recognition across pose.

Learning Object Representations Using A Priori Constraints Within ORASSYLL

In this article, a biologically plausible and efficient object recognition system (called ORASSYLL) is introduced, based on a set of a priori constraints motivated by findings of developmental psychology and neuro-physiology. These constraints are concerned with the organization of the input in local and corresponding entities, the interpretation of the input by its transformation in a highly structured feature space, and the evaluation of features extracted from an image sequence by statistical evaluation criteria. In the context of the bias-variance dilemma, the functional role of a priori knowledge within ORASSYLL is discussed. In contrast to systems in which object representations are defined manually, the introduced constraints allow an autonomous learning from complex scenes.

Information Theory and the Brain: Principles of Cortical Processing Applied to and Motivated by Artificial Object Recognition

In this paper we discuss the biological plausibility of the object recognition system described in detail in (Kruger, Peters and v.d. Malsburg, 1996). We claim that this system realizes the following principles of cortical processing: hierarchical processing, sparse coding, and ordered arrangement of features. Furthermore, our feature selection is motivated by response properties of neurons in striate cortex and by Biederman’s theory of object representation on higher stages of visual processing (Biederman, 1987). Inspired by the current discussion about aspects of cortical processing, we hope to derive more efficient algorithms. By discussing the functional meaning of these aspects in our object recognition system, we hope to attain a deeper understanding of their meaning for brain processing.

Learning Weights in Discrimination Functions Using a priori Constraints

We introduce a learning algorithm for the weights in a very common class of discrimination functions usually cailed “weighted average”. Different submodules are produced by some feature extraction and are weighted according to their significance for the actual discrimination task. The learning algorithm can reduce the number of free variables by simple but effective a prion criteria about significant features. We apply our algorithm to three different tasks all concerned with face recognition: a 40 dimensional and an 1800 dimensional problem in face discrimination, and a 42 dimensional problem in pose estimation. For the first and second task, the same weights are applied to the discrimination of all classes; for the third problem, a metric for every class is learned. For all tasks significant improvements could be achieved. In the third task the performance was increased from 80% to 90%. The idea of our algorithm is so general that it can be applied to improve a large number of existing pattern recognition systems.