SAR Image Change Detection method of MDS-SRM Based on Hybrid Cascade Merging

Abstract

Since the advantage of independence of data probability distribution and anti-noise performance, SRM (Statistical Region Merging) is widely applied in image segmentation field. But only the gray feature of the image has been considered in SRM's sorting process and lacked the sorting optimization process, which leads to false contour phenomenon caused by local over-merging (less segmentation) in the image merging process. In this paper, an improved SRM algorithm named MDS (Multilayer Dynamic Sorting)-SRM is presented for SAR image change detection, more features include gray mean difference, adjacency correlation and the area difference of merging areas are considered to dynamic sorting to overcome the over-merging in MDS-SRM. Firstly, non-local mean filter is used to remove the speckle noise of SAR image, the double-channel difference image are obtained by logarithmic ratio algorithm and mean ratio algorithm; Secondly, a three-level hybrid cascade structure for SAR image change detection has been designed. In the process of first stage original SRM is employed to primarily merge the difference images, the second stage uses an improved MDS (Multilayer Dynamic Sorting) -SRM method to execute the secondary merging, in third stage a simplified SRM is utilized to combine the merging result to changed part and unchanged part. The experiment results showed that our method can obtain the better detection performance than that of the original SRM and RFLICM algorithm.