Image registration based on the feature extraction is widely used in the medical field. In order to apply the scaleinvariant feature transform algorithm into feature extraction and matching of vascular images, a suitable low contrast threshold is determined with the curve fitting method according to the vascular image characteristic, besides, in order to improve the speed and matching accuracy of the SIFT algorithm, conduct dimensionality reduction on the feature descriptors of SIFT algorithm. A method based upon the modulus comparison was used for feature points matching, find the matching points by comparing the relationship between the feature points vector modulus. The experimental results and data show that the improved algorithm has increased the performance in improving the matching rate and lower false matching rate,and it is important for the clinical treatment of vascular disease.