A neural network model for the mechanism of selective attention in visual pattern recognition
Kunihiko Fukushima · Systems and Computers in Japan · 1987
Abstract When a complex figure consisting of two or more patterns is presented, we, as human beings, can selectively concentrate on these patterns one at a time, and recognize individual patterns in turn. At the same time, we can separate a component of the pattern being recognized, and extract it from the rest of the figure. Even if one of the patterns on which we are concentrating contains noise or defects, we can recall a complete pattern from which the noise and the defects have been eliminated. It is not necessary for perfect recall that the stimulus pattern be identical in shape to the pattern which we learned. Even though the pattern is distorted in shape or changed in size, we can recognize it and eliminate defects by interpolation. During the process of interpolation, we make full use of even slight traces in the defective parts of the pattern on which we are selectively concentrating, and recall the perfect original pattern. A model which performs this function of the human brain, that is, of the function of the selective concentration in visual pattern recognition, is proposed, and its behavior is demonstrated by computer simulation. The model consists of a hierarchical neural network which has efferent connections between cells.