In this paper, we propose a novel and efficient method for detecting distribution changes in a picture. We design an approach based on the parametric change-points detection problem and apply it into region detection problem. The number of change-points is determined by the Schwartz information criterion and the location of the change-points is estimated via a modified pruned exact linear time algorithm. Furthermore, we propose a new merging method to merge the regions in the same distribution after finishing the estimation of change-points. Under some mild conditions, we prove our estimation of change-points and regions is consistent. In the simulation studies, the new method performs well in both estimation and computation time compared to other method.
LI Jiaqi,LI Zhonghua,WANG Xiaopu. An Efficient Approach to Detect Changed Regions in the Image[J]. CHINESE JOURNAL OF APPLIED PROBABILITY AND STATIST, 2020, 36(3): 295-320.