Cognitive Vision Inspired Contour and Vertex Detection

Abstract

This paper presents a visual cortex inspired cognitive model for contour and vertex detection. The model is strongly based on the receptive field characteristics of cortical neurons of the visual cortex. As a step forward compared to the previous version of the model, a new dimension has been added, which replaces the binary signals and operations by operations on real values. The resulting system yields a better approximation of the biological system, as well as provides stronger and more distinct contour lines and vertices. The contour detection and vertex extraction is performed by a vast network of simple units of computation simultaneously processing the visual data. The computational units are organized in a special structure, the Visual Feature Array (VFA), which allows the structural representation of complex operations. The goal of the model is to extract abstract information from an image, which in turn may be used as input for the recognition process of even more abstract visual objects. In order to achieve constant time execution of the model, the aspects of hardware implementation are also treated in this paper.