Color image segmentation based on watershed transformation upon homogeneity image and two-step region merging
基于均匀性图分水岭变换及两步区域合并的彩色图像分割

A novel algorithm based on watershed for color image segmentation was presented. The original color image was initially smoothed using peer group filtering in order to preserve the location of the edges. A homogeneity image was defined, which integrated gradient with local second-order moment information, rendered more local information, and then was segmented using traditional watershed algorithm to yield mosaic image. A two-step region merging was carried out aeeording to region area and Fisher distance. Color diversion increment deduced from relationship between information lost and the number of regions retained was adopted as automatic terminate strategy to refine the result of segmentation. Experiment results show the effectiveness of the method.