Seamless Stitching of Panoramic Image Based on Multiple Homography Matrix

Abstract

Because the traditional method of image stitching based on feature point detection tends to fracture in the stitching results, As Projective-As Possible (APAP) stitching algorithm proposed by Juilo Zaragoza was improved in this paper to eliminate fracture in stitching overlap area. First of all, the matching feature points were calculated through A-KAZE feature point detection algorithm. Then, the projection model was determined on the basis of Direct Linear Transformation (DLT) algorithm, and the target images were divided into 25×25 blocks, which were projected with APAP algorithm. Besides, multiple homography matrix was calculated. Finally, the global projection transformation model was estimated according to each multiple homography matrix. Considering the traditional image fusion approach makes stitching results discontinuous, and cannot guarantee the integrity of the objects in image overlap, the Min-Cut/Max-Flow of edge detection operator and Laplacian multiresolution fusion algorithm were added in this paper so as to achieve image fusion and realize the seamless stitching of multiple images. The simulation experiments show that the stitching method can eliminate the fracture in the stitching results, and realize the seamless stitching of multiple images.