In order to realize quickly and accurately matching between the image features, an efficient high-dimensional feature vector retrieval algorithm, Extended Locality Sensitive Hashing(ELSH), was proposed based on LSH(Locality Sensitive Hashing). Firstly, the Scale Invariant Feature Transform (SIFT) algorithm was used to get the special point of an image and its features. Then, according to the sub-vectors selected randomly from the SIFT features, a hash index structure was built to reduce the indexing dimension and the searching scope. Thus, it can significantly reduce the time cost of indexing. Finally, the Random Sample Consensus (RANSAC) algorithm was used to select the right feature point pairs. Experimental results indicate that compared with the Best-Bin-First(BBF) and the LSH algorithm, ELSH algorithm not only ensures the accuracy of matching points, but also reduces the matching time. The time cost of ELSH only takes 50.1% of that of the BBF, and 62.1% of that of the LSH. In conclusion, the proposed algorithm can quickly and precisely achieve the registration between images.