The registration of multitemperal remote sensing images before and after disaster is a challenging problem.The main reasons include: (1) The area-based registration methods are difficult for the task because the graylevel of remote sensing images usually varies.(2) The feature-based registration methods usually can't work well, because of the lack of honologus control points, which is caused by the damage of salient features in the scene such as roads and bridges.The method of manual control points selection for registration of remote sensing images, which have large data set, is very time consuming.