SIFT mismatching points eliminating algorithm based on region overlapping kernel weighted Hu moment

A kind of scale invariant feature transform (SIFT) mismatching points eliminating algorithm based on region overlapping kernel weighted Hu moment is proposed to solve the mismatching problem of SIFT. Firstly, this algorithm computes Hu moment overlap in the 4-neighborhood of SIFT descriptors, which can generate the seed point descriptor with contour feature and texture feature. Then according to the characteristic of the SIFT region, this algorithm uses a kernel function to weighting the seed point to generate a 63-dimensional feature point descriptor. Finally, the Bhattacharyya coefficient is used to compute similarity of matching points, and eliminates those matching points with low similarity. The proposed algorithm is compared with three other algorithms, the experimental results show that the proposed algorithm has the best robustness and real-time feature, and a good comprehensive performance also.