Region Merging with Texture Pattern for Segmentation of SAR Images

Abstract

This paper proposes a novel merging cost with texture pattern discrimination for region merging in the segmentation of SAR images, which integrates texture patterns, statistical characteristics, and shape prior of SAR images. By the initial partition based on multiscale Bhattacharyya distance, the input image is first over-segmented into region patches. Then, in the region merging process, a correlation measure of texture patterns between adjacent patches is constructed to obtain the proposed merging cost by fusing it with statistical similarity measure (SSM) and the relative common boundary length penalty (RCBLP) according to the pixel number of region patches. Finally, the segmentation result is output by iteratively merging the pair of adjacent regions with the smallest merging cost until the end condition is satisfied. Experimental results on real SAR images indicate that the proposed method is competitive in the segmentation of SAR images with complex scenes in comparison with several recent state-of-the-art ones.