Chun-Yuan Wu, Yung-Sheng Chen and Wen-Hsing Hsu+
Institute of Electrical Engineering
National Tsing Hua University
Hsin-Chu, Taiwan, 30043, R.O.C..
+Institute of Information Science
Nankang, Taiwan, 11529, R.O.C.
In this paper, we present a new algorithm to simulate thick-line pattern recognition, which is based on line continuation. Line, continuation is a fundamental intuitive property in human visual perception. This property is much related to the hman visual system, especially at the area of simple cells in visual cortex. In addition, this property is suitably applied to recognition of thick-line patterns, due to the tolerance range of the visual receptive field The proposed algorithm basically consists of three modules: directionization, characterization, and figurization. To tolerate a certain extent of deformation, we first convert the input pattern into a map of direction codes through module 1, extract the direction codes, which are the cross references of the skeleton of the input pattern, by module 2, and then complete the matching operation by module 3. The matching mechanism of module 3 is concerned both with the pattern and background of the stimulus. Due to the tolerance of limited character deformations of human visual perception, a matching weight, called "close relation" is proposed. We apply the algorithm to a numeral("0" to "9") recognition system. The result of the simulation is listed and analyzed. Finally, the characteristics of this model are commented on, and the influence of noise is also discussed.
Keywords: visual perception, continuation, global viewing, figure-and-background, autofocus, human behavior in walking (HBW), close relation
Received December 15, 1989; revised July 15, 1990.
Communicated by J. H. Huang.