Based on Statistics of the Gradients the Feature Matching Algorithm

The feature matching is the first step of several computer vision duties. In this paper we provide a new feature detect and matching approach based on statistics of the gradients of the feature region. It is extension of the sift algorithm. The algorithm represented in this paper can be used to perform reliable matching to image sequence, which have larger change in 3D viewpoint and change in illumination. The new approach not only describes the local region structure but also statistic orientation histogram of periphery twelve 5X5 sample sub-region. The constructed descriptor has more robust distinguishability to support the following match step. In the matching step, by using polar, symmetry and uniqueness constraints to filter the tentative feature pairs, many outliers are eliminated and the correct feature pairs are obtained in the last step. By comparing with traditional Harris--Correlation matching algorithm as well as theoretical analysis, the algorithm that this article proposed has the higher stability as well as the anti-jamming.