A Lunar Impact Crater Model is proposed by analyzing the relationship between region of interest (ROI) and local salient features of lunar imagery according to the illumination characteristics. Based on the model, an ROI detection algorithm is proposed. SURF of the highlight region and shadow region of an impact crater are extracted at first; then the homogeneous features are merged into new ones. According to the constructive constrains of the Lunar Impact Crater Model, pairs of highlight feature and shadow feature are combined to generate the ROI. The algorithm has been put into test on Chang’e-1 and Chang’e-2 lunar imagery data, and was able to correctly detect the obvious regions of impact craters with results much better than those obtained by Itti algorithm.