Research Of Image Correspondence Based On Geometry Constraint

Image correspondence is a very hot topic for image panorama, remote sensing and image retrieval. The motivation of image correspondence intends to measure how well one image match another one.This paper proposes two approaches for image correspondence with geometrical constraints in order to prevent error propagation during image correspondence.First, this paper st udies t hose v isual f eatures w hich have be en us ually implemented in image correspondence including multi-scale features.Second, two approaches for image correspondences with geometrical constraints are put forth i n t his pa per, na mely t riangle c onstraint and gr aph c onstraint. Image correspondence w ith triangle cons traint de tects t he t riangles t hat m ay have be en mismatched or located in changed regions and assigns those triangles with a lower priority. Therefore, it is effective to control the error propagation during the matching process. Image correspondence with graph constraint introduces a voting strategy to improve the matching result.Finally, this paper presents a r emote sensing system. The issues such as image correspondence, clustering, segmentation and vectorization involved in remote sensing are discussed.