Finding Correspondences Between Images using Descriptors and Graphs

AbstractThe problem of finding correspondences is considered in the article. The main objective of this method is to reduce the number of false matches by using structural performance. The relevance of the problem is proven. The review of existing methods of finding correspondences is provided. The method presented is finding correspondences based on combined use of graphs and descriptors. Scott and Longuet-Higgins algorithm is used in the first stage. We construct a graph the vertices of which are the features on the two images. Singular value decomposition of the graph matrix is performed. The correspondences based on the descriptor are used. An example of the algorithm is shown. Test images are researched. A comparison of the algorithm with the RANSAC is carried out. The proposed approach allows excluding a significant portion of false correspondences found using the existing descriptors. The algorithm has high speed