Adaptive clustering feature matching algorithm based on SIFT and RANSAC

Scale-invariant feature transform (SIFT) algorithm is widely used in the field of image matching. The traditional SIFT does not perform well in terms of computational complexity and matching accuracy in complex scenes. Based on the above problems, an improved SIFT and RANSAC algorithm is proposed in this paper. When selecting extreme points, the original 26 reference points in DOG space are extended to 32 reference points within Manhattan distance 2. By establishing 8 concentric circle neighborhoods with increasing radius in the sampling sub area, the latitude of keypoint descriptor is reduced to 64 dimensions. When performing matching, it first goes through KNN coarse matching, and then uses the score function to calculate the scores of the two clustering result sets, keypoint descriptor Euclidean distance clustering and spatial location clustering, and uses the high score set as input to improve the input parameters and the number of iterations of RANSAC. The simulation results show that the algorithm in this paper can further improve the correct matching rate with good robustness in a variety of scenes compared with the traditional algorithm and some other related algorithms.