A cascaded segmentation method based on region merging to change detection in remote sensing images

Abstract

Change detection based on image superpixels can extract more geomorphologic information among multitemporal remote sensing images than methods based on pixel difference. In this paper, we presented a cascaded segmentation method to extract clear change region boundry with noise supperssion. First, Simple linear iterative clustering (SLIC) is used to generate super pixels which adhere difference image boundries tightly for purpose of searching change regions. Second, one Statistical Region Merging (SRM) with dynamic sorting algorithm is modified to merge those homogeneous super pixels. After the candidate change regions established, classified change map are remerged by using simplified SRM. Finally, the proposed method are compared with methods based on PCA and MRF. Experimental results shows our method restrain the over segmentation and obtain better performance of change detection than conventional SRM algorithms.