Abstract: In order to solve the over-segmentation problem in the image segmentation,a graph-based color image segmentation algorithm is proposed.In the algorithm,on the basis of the region-merging method,the over-segmentation regions are obtained by using Mean shift to preprocess an image,and for the over-segmentation regions,a region adjacency graph is constructed.Then,the color,texture and edge contour similarities between adjacency regions are measured to judge if adjacency regions need to be merged,until all the satisfactory regions are merged.Besides,the regions are merged by searching the optimal merging-cost,so as to preserve some global prosperity of the image.Experimental results indicate that the proposed algorithm can completely segment object regions even when the color features between background regions and object regions of an image are similar,and it has a better segmentation performance in comparison with four state-of-the-art segmentation algorithms.

Key words: image segmentation, Gabor filters, textures, edge, regional similarity