An Iterative Algorithm for Finding Point Correspondences

This paper presents a solution to a general correspondence problem between a set of points (or image features) and a template, where the matching criterion includes linear parameters which reflect dynamic camera zooming during image-based tracking procedures. An algorithm is proposed which will solve this type of correspondence problem. Unlike most existing algorithms, which approach the solution using relaxation and mathematical programming, the proposed algorithm searches for the solution by iteratively interchanging the position of feature-pairs within the feature set until a necessary condition is satisfied. This being, if the interchange of two features from the feature set degrades the matching criterion. Three sets of correspondence examples are detailed which illustrate the effectiveness and efficiency of the proposed correspondence technique in solving cases, or image data, which are affected by large translation and scaling. A standard technique, the Scott and Longuet-Higgins method, taken from literature is also tested as a comparison.