FIGURE SEGREGATION MODEL USING STOCHASTIC RELAXATION METHOD

In its psychology, figure segregation is a visual function of the huma n being which regards an overlapped figure as being composed of severa l divided subfigures. To construct computational figure segregation mo dels, two processes are needed: the selection of a few available candi dates for segregation among all. conceivable candidates, and the estim ation of the selection rate of these candidates, Few models including the two forementioned processes have been proposed. As a result, a fig ure segregation model is proposed that enables both processes to be pe rformed using a stochastic relaxation method in statistical thermodyna mics. This approach has two advantages: knowledge about the visual fun ction can be embedded easily into the method as energy functions; and solutions can be obtained with probability that can be a measure of se lection. As a result of computer simulation for 10 different figures, 96.8 percent of all candidates which are extracted by 100 subjects are extracted correctly, and the selection rate of the candidates is very close to that in human experiments.