Homogeneous region segmentation for SAR images based on two steps segmentation algorithm

This paper presents an unsupervised algorithm for the segmentation of SAR images homogeneous regions, based on a likelihood difference function, which is derived from the Gamma distribution. The idea behind the proposed algorithm is to fuse the over-segmented image blocks, according to a similarity criterion. The proposed schema includes two steps segmentation, a rough segmentation and a fine segmentation. In the rough segmentation, image blocks are fused to form a large region, in which a likelihood difference function based on Gamma distribution is employed as a similarity measure. In the fine segmentation, a pixel-level segmentation is carried out, aiming to obtain a more smooth segmentation edge. Experimental results on two SAR image are given to demonstrate the effectiveness of the proposed scheme.