A Model for Visual Feature Extraction Based on the Mammalian Visual Cortex

Abstract

The present paper proposes a model for intelligent image contour detection. The model is strongly based on the architecture and functionality of the mammalian visual cortex. A pixel-to-feature transformation is performed on the input image as the afferent visual information. The result of the transformation is a three-dimensional array of data representing abstract image features (contour objects), instead of another array of pixels. The contour feature recognition is performed by a vast and complex network of simple units of computation that work together in a parallel way. The use of a large number of such simple units allows a clear structure that can be implemented on a special hardware to allow fast, constant time feature recognition.