Image Mosaic using ORB descriptor and improved blending algorithm

Abstract

This paper presents an image mosaic technique using ORB (oriented FAST and rotated BRIEF) descriptor and an improved blending algorithm. Features of images with overlapping regions are first described by ORB descriptors. This binary descriptor outperforms SIFT and SURF in regard of processing speed. RANSAC algorithm is then applied to reject mismatches. Once the best projective deformation matrix is found, brightness equalization is conducted and an improved weighting function is utilized to blend images and reduce visible artifacts. Experimental results show this approach to be efficient in accounting for image mosaic, and the edge artifacts could be significantly reduced and even completely eliminated with the blending process designed in this paper.