The problem of establishing correspondences between images taken from different viewpoints is fundamental in computer vision. We propose an algorithm which is capable of handling larger changes in viewpoint than classical correlation based techniques. Optimal performance for the algorithm is achieved for textured objects which are locally planar in at least one direction. The algorithm works by computing affinely invariant fourier features from intensity profiles in each image. The intensity profiles are extracted from the image data between randomly selected pairs of image interest points. Using a voting scheme, pairs of interest points are matched across images by comparing vectors of fourier features. Outliers among the matches are rejected in two stages, a fast stage using novel view consistency constraints, and a second, slower stage using RANSAC and fundamental matrix computation. In order to demonstrate the quality of the results, the algorithm is tested on several different image pairs.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.