An Image Stitching Algorithm Based on SIFT and DAISY Descriptor

Abstract

Correspondences matching is essential to image stitching and greatly influences the stitching quality as the homography matrix is calculated from the correspondences. Mismatching would be generated by using SIFT alone when dealing images with repeat similar structures. To improve the matching accuracy, SIFT and DAISY are combined to extract and match feature points. Then the homography matrix is computed by least square, RANSAC and bundle adjust methods. Experiments show that the matching accuracy is improved and the stitching results are well and robust.