Image registration is an important subject in computer vision. How to improve the probability of registration and precision of localization is a hard task. A product correlation algorithm of image registration based on feature and spatial-temporal relationship is put forward. Firstly, extracting edge features of image using wavelet for sequential real-time images and base image. Secondly, the sequential real-time images (3 frame) are matched with base image using normalized cross-correlation, and the registration points searched in sequential correlation matrix are fused according to the offsets. This method overcomes the shortcoming that the probability of registration is decreased with area increasing because of geometric distortion, and improves the robustness of registration.