A probabilistic approach is presented for fast CCD image matching.Firstly,the lunar contour feature is constructed based on lunar component models and a full shape model.Secondly,the critical points of the contours are extracted as feature points and then a subset of feature points are randomly selected to instantiate a candidate transform,followed by a process of validating and refining the transform parameters.Experiment results show that the approach can match two images in linear time,which is about one order of magnitude faster than the traditional method.