In this paper we present an effective algorithm for automatic image registration by matching features in images made from different viewpoint. For the SIFT detector can assure local variant of image features such as translation, scaling and rotation, we use SIFT to implement the image registration. But the SIFT usually bring too many matching points and outliers removing process is needed. We present an SIFT based algorithm which get rid of redundant matching points by an estimated threshold from multiple experiments. From the experiments, we found our algorithm produce much less match points and the correctness rate increased significantly.