A novel similarity-based approach for image segmentation

In this paper we propose a similarity-based approach which segments an image into different regions according to a similarity measurement C which reflects the relationship between a pixel and its neighborhood. We assume that each pixel in the same region should have a similar relationship with its neighborhood. On the basis of this assumption, we divide the range of C into several sub-ranges. Those pixels whose Cs belong to the same sub-range would be segmented to the same region. Furthermore, in order to process those pictures with too much detailed information which could affect the result of segmentation, we present an information-lessening method to reduce unnecessary details. Experimental results have demonstrated that the novel segmentation approach can work effectively.