Observing damages immediately by detecting changes from geographic images is required when large scales of disasters occur. In our previous work, we proposed an unique adaptive nonlinear stereo mapping method, which compares aerial images, or satellite images taken at a different time, and have realized the accurate change detection in the case with parallel translations between two images. In this paper, we examine the method, based on the previous one, which realizes the accurate change detection from images including rotational translations or nonlinear distortions. The multi-leveled exploration process simplifies parameter settings and realizes the efficient exploration of corresponding areas. Further more, correct transfer vectors can be constructed by the selective smoothing, which replaces transfer vectors with a neighborhood vector selectively. By evaluating changes detected by the proposed method, with ROC (Receiver Operating Characteristic) method, the proposed method is proved to be in the stable accurate change detection compared with the previous method.
View full abstract