Image Stitching Based on Local Symmetry Features

Traditional image stitching methods represented by SIFT are sensitive to non-linear illumination changes. In this paper, a new algorithm is presented for image stitching based on local symmetry features. Firstly, feature points are extracted using the detector based on local symmetry. Secondly, SIFT descriptor and local symmetry descriptor are combined to characterize those feature points. Thirdly, feature matching is carried out by randomized KD-trees and transform parameters are calculated by the correct inner points after the RANSAC was used to eliminate wrong matches. Finally, image stitching is completed with smoothing algorithm. The experimental results indicate that the proposed method has a higher matching precision than SIFT and SURF under the non-linear illumination change scenarios and can achieve better performance in image stitching.