Self-Organization of Target Recognition Based on a Unified Cortical Model-Based Foveation Algorithm

摘要

This paper proposes an artificial target recognition (ATR) system that can efficiently recognize object features in image features, particularly the eyes from different individuals, in different geometrical views and environmental conditions. Motivated by the applications of foveated digital images, in humans visualization, a new ATR system is presented. Humans recognize visual object features by foveation. The eyes accumulate different coordinates about the image feature to identify the object feature within. The locations of the foveation coordinates are usually the corners, and some of the times, the edges of the object feature within the image feature. The unified cortical model (UCM) has the inherent ability to segment images. The pulse images generated from the image feature by the UCM are corners and edges which are similar to the foveation coordinates. The Gabor wavelet filters are used couples with the generated foveation coordinates to extract jets or vectors in the image feature plane. The self-organizing maps (SOM) is used to train the Gabor jets for object features recognition. Some sample calculations using the UCM are also presented.