Several problems in feature matching are studied in this dissertation, which involve feature point matching based on local feature descriptor, feature point matching of images containing repetitive patterns and line matching. The main contributions and novelties are the following: (1) Based on the experimental analysis of the disadvantages of existing local feature descriptors, an intensity-order based local feature pooling strategy is presented. By pooling gradient-based features and intensity-based features respectively, two kinds of local feature descriptors are obtained: MROGH and MRRID. Experimental results on image matching and object recognition demonstrate their effectiveness and superiority to other local descriptors in the literature. (2) In order to establish reliable point correspondences between images containing repetitive patterns, a novel method based on pairs of interest points is presented. It starts from matching pairs of interest points and then establishes point correspondences from the matched point-pairs on the basis of geometric consistency of corresponding points. Its effectiveness for matching images of repetitive patterns and its superiority to other methods proposed in the literature are demonstrated by experimental results. (3) Based on the introduced line-point invariants, a novel line matching method is presented, which conducts line matching by the help of the result of feature point matching. Although the proposed method is based on the result of feature point matching, it is robust to mismatches in point correspondences. Experimental results on various image transformations show its effectiveness as well as its superiority to the state-of-the-art methods.
修改评论