Nikolay Petkov

Nonlinear operator for oriented texture

Texture is an important part of the visual world of animals and humans and their visual systems successfully detect, discriminate, and segment texture. Relatively recently progress was made concerning structures in the brain that are presumably responsible for texture processing. Neurophysiologists reported on the discovery of a new type of orientation selective neuron in areas V1 and V2 of the visual cortex of monkeys which they called grating cells. Such cells respond vigorously to a grating of bars of appropriate orientation, position and periodicity. In contrast to other orientation selective cells, grating cells respond very weakly or not at all to single bars which do not make part of a grating. Elsewhere we proposed a nonlinear model of this type of cell and demonstrated the advantages of grating cells with respect to the separation of texture and form information. In this paper, we use grating cell operators to obtain features and compare these operators in texture analysis tasks with commonly used feature extracting operators such as Gabor-energy and co-occurrence matrix operators. For a quantitative comparison of the discrimination properties of the concerned operators a new method is proposed which is based on the Fisher (1923) linear discriminant and the Fisher criterion. The operators are also qualitatively compared with respect to their ability to separate texture from form information and their suitability for texture segmentation.

Biologically motivated computationally intensive approaches to image pattern recognition

This paper presents some of the research activities of the research group in vision as a grand challenge problem whose solution is estimated to need the power of Tflop/s computers and for which computational methods have yet to be developed. The concerned approaches are biologically motivated, in that we try to mimic and use mechanisms employed by natural vision systems, more specifically the visual system of primates. Visual information representations which are motivated by the function of the primary visual cortex, more specifically by the function of so-called simple cells, are computed. Three different methods for using such representations to solve image pattern recognition problems are presented. These are: (i) extraction and comparison of lower-dimension representations, (ii) computing optimal mappings of an image onto other images by optic flow techniques and (iii) application of a self-organising neural network classifier. The problems of automatic recognition and classification of visual patterns, in particular the discrimination of human faces, are used to test the usefulness and feasibility of these approaches.

A Computational Model of Periodic-Pattern-Selective Cells

A computational model of so-called grating cells is proposed. These cells, found in areas V1 and V2 of the visual cortex of monkeys, respond strongly to bar gratings of a given orientation and periodicity but very weakly or not at all to single bars. This non-linear behavior is quite different from the spatial frequency filtering behavior exhibited by the other types of orientation selective cells. It is incorporated in the proposed model by using an AND-like non-linearity to combine the responses of simple cells and compute the activities of so-called grating subunits which are subsequently summed up. The parameters of the model are adjusted to reproduce the results measured by neurophysiologists with different visual stimuli. The proposed computational model of a grating cell is used to compute the collective activation of sets of such cells, referred to as cortical images, induced by natural visual stimuli. On the basis of the results of such simulations we speculate about the possible role of grating cells in the visual system and demonstrate the usefulness of grating cell operators for some computer vision tasks, such as automatic face recognition and document processing.

Person identification based on multiscale matching of cortical images

A set of so-called cortical images, motivated by the function of simple cells in the primary visual cortex of mammals, is computed from each of two input images and an image pyramid is constructed for each cortical image. The two sets of cortical image pyramids are matched synchronously and an optimal mapping of the one image onto the other image is determined. The method was implemented on the Connection Machine CM-5 of the University of Groningen in the data-parallel programming model and applied to the problem of face recognition.

Biologically Motivated Approach to Face Recognition

A biologically motivated compute intensive approach to computer vision is developed and applied to the problem of face recognition. The approach is based on the use of two-dimensional Gabor functions that fit the receptive fields of simple cells in the primary visual cortex of mammals. A descriptor set that is robust against translations is extracted by a global reduction operation and used for a search in an image database. The method was applied on a database of 205 face images of 30 persons and a recognition rate of 94% was achieved.

Image classification system based on cortical representations and unsupervised neural network learning

A preprocessor based on a computational model of simple cells in the mammalian primary visual cortex is combined with a self-organising artificial neural network classifier. After learning with a sequence of input images, the output units of the system turn out to correspond to classes of input images and this correspondence follows closely human perception. In particular, groups of output units which are selective for images of human faces emerge. In this respect the output units mimic the behaviour of face selective cells that have been found in the inferior temporal cortex of primates. The system is capable of memorising image patterns, building autonomously its own internal representations, and correctly classifying new patterns without using any a priori model of the visual world.

Use of Cortical Filters and Neural Networks in a Self-Organising Image Classification System

A preprocessor based on a computational model of simple cells in the mammalian primary visual cortex is combined with a self-organising artificial neural network in order to to set up an image classification system. After learning with a sequence of input images, the output units of the system turn out to correspond to classes of input images. Notably, a group of output units which are selective for images of human faces emerges. These units mimic the behaviour of face selective cells that have been found in the inferior temporal cortex of primates.

Utilizing fixed-size systolic arrays for large computational problems

A new scheme for partitioning systolic algorithms is presented. It is based on the time-sharing properties of the c-slow circuits. The technique is amenable to formalization and holds high potential for automatization.

Design of Bit-level systolic convolvers for image processing

A method for the design of bit-level systolic FIR filters with constant coefficients is presented. A convolution is represented as a sum of convolutions with coefficients which are powers of 2. It is then implemented in a two-dimensional bit-level systolic array of full adders. The worst-case space-time requirements of the algorithm involved are superior to the space-time requirements of the systolic convolution algorithms previously known.