Abstract:

       A self-organizing map of local parameters is proposed to simulate the initial visual processing mechanism for color scenes. The method employs local color mean values, local color contrasts, and positions as features, and uses a self-organizing map for classification to deal with multiple tasks such as image segmentation, edge detection, and contour separation. Experimental results for both synthetic and natural color images show that the method can perform these tasks simultaneously. It maps image parts with different features to different positions of the output space, conforming to the feature mapping and simultaneously processing characteristics of the visual system. This method can be used to investigate and simulate visual mechanisms, and handle multiple image processing tasks.

Key words: visual mechanism simulation , self-organizing map , image segmentation, edge detection , contour separation