Abstract:

In the matching of image features, a great deal of peaks may occur and unreliable and inaccurate matching results are obtained due to the gray measure. In order to solve these problems, a new matching method of image features is proposed based on the moment invariants of rotation, scale, translation and contrast （RSTC）. In the new matching method, an improved SUSAN algorithm is employed to detect the comers. Then, a new moment invariant of RSTC is constructed to describe the corner features and to measure the similarity of corner matching. Moreover, a guided matching with RANSAC robust estimation and epipolar line constraint is performed. Experimental results of actual images show that the proposed method eliminates the long lines caused by wrong feature matching and is of excellent matching effectiveness, and that it improves the matching precision by more than 6% as compared with the gray matching method.

Key words: image matching, feature extraction, SUSAN algorithm, comer detection, moment invariant, robust estimation, epipolar line