An improved sift image matching detection

Ning Liu, Xiaoyu Zhang 

Department of Computer Science and Applications, Zhengzhou Institute of Aeronautical Industry Management, China

Aiming at the problems - edge response in the traditional SIFT descriptor and the insufficient correct matching feature points, the work proposed a kind of improved SIFT Image Matching Detection Algorithm. The candidate key point was firstly detected by the SIFT algorithm; Canny edge detection algorithm was used to detect image edge points; it was judged whether the candidate key point needed to be eradicated by comparing whether the candidate key point equals to the coordinates of edge points; K-means clustering pattern, which is combined by the vector space cosine similarity and vector Euclidean distance similarity, was adopted to perform global image similarity matching. Finally, RANSAC algorithm was used to further get rid of the wrong matching. The experimental result indicates the improved method greatly enhances the stability of SIFT algorithmic and the accuracy rate of matching.