Research On Image Matching Based On Point Pattern Matching Of Graph Theory

Image matching is an important issue in image processing, which is a foundational problem in computer vision theory and application. The results of its research have widely and practically application background in many areas. The matching based on local invariant feature can reduce time complexity and improve the precision of image matching. Therefore, point pattern matching can be completed using point features in image with the space transformation parameters or invariant. Point pattern matching based on graph theory is a hot and difficult research branch, which has made great progress these years. This dissertation is focused on point pattern matching based on graph theory. The main contributions of the dissertation are summarized in the following.(1) The paper presents the scientific signification and application value of point pattern matching, reviews the related research status and the future development of point pattern matching based on graph theory. The crucial research contents of the dissertation are pointed out.(2) As the SIFT operator may extract more false keypoints in the image with various texture, which would affect the result of image matching. This paper propose a new algorithm of image matching based on SIFT local invariant feature of Harris threshold criterion. On the basis of extracting SIFT invariant features, the extracted invariant feature is selected based on Harris threshold criterion. Therefore, there leaves some more robust and well separable features because the worse separable features are rejected in some region of close-grained image.Finally, the vector of invariant feature and Graph Transformation Matching method is used to match accurately.(3) A novel and robust point pattern matching algorithm based on Quasi Laplacian Spectrum and Point Pair Topological Characteristic (QLS-PPTC) is proposed. Firstly, a Signless Laplacian matrix is constructed by using the minimal spanning tree of weighted graph, and then the eigenvalues and eigenvectors obtained from the spectrum decomposition are used to represent the point’s features, which make it possible to calculate the matching probability. Secondly, the comparability measurement of point pair topological characteristic is computed to define local compatibility between the point pairs, and the correct matching results are achieved by using the method of probabilistic relaxation. The contrast experimental results show that the proposed algorithm is robust when the outliers and noises exist in point matching.(4) A point pattern matching algorithm based on Mahalanobis distance is proposed, which effect is analyzed and confirmed by experiments. Secondly, the Graph Transformation Matching algorithm and Weighted Graph Transformation Matching algorithm are studied deeply. To overcome the limitation of Mahalanobis distance and WGTM, a novel and robust point pattern matching algorithm based on Weighted Graph Transformation using Mahalanobis distance is proposed. The similarity evaluated by Mahalanobis distance is embedded into WGTM algorithm under the constraint of median distance and angular distance. Then point pairs are obtained through iteratively eliminating the outliers. Experimental results on synthetic data and real-world data demonstrate that the proposed algorithm is effective and robust.