Subdivision Grid Combined With Content Perception Large Parallax Image Stitching Algorithm

Aiming at the problem of ghosting in overlapping regions and severe distortion in non-overlapping regions of large parallax image stitching algorithm,a large parallax image stitching algorithm based on segmentation grid combined with content perception is proposed.Firstly,the SIFT algorithm is used to perform feature point matching on the two images,and the random matching consistency algorithm is used to screen out the optimal matching points.Then,according to the distribution of the paired matching points,the grid of the projection algorithm is segmented as much as possible,and each network is calculated.The homography matrix of the lattice achieves the purpose of precise registration of overlapping regions;then,the global optimal similarity transformation matrix is selected by the random sampling consistency algorithm,and the grid homography matrix is weighted and superimposed with the global optimal similarity transformation matrix.Reduce the perspective distortion of the non-overlapping area;finally,the content perception of the overlapping area,find the best stitching,remove the higher importance area,and retain the lower importance area for splicing,which increases the sharpness of the overlapping part of the stitched image.In this paper,the algorithm as far as possible projection(APAP)algorithm,semi-projection transformation(SPHP)algorithm,as far as possible natural projection(AANAP)algorithm on the public data set for visual sensory evaluation,overlap region error root mean square and splicing efficiency The experimental results show that the proposed algorithm not only effectively prevents overlapping regions from appearing,but also reduces the warpage of non-overlapping regions.The error root mean square of overlapping regions corresponding to stitched pixels is reduced by 40%compared with APAP algorithm.On the left and right,compared with the SPHP algorithm,the average reduction is about 60%,and the average reduction is about 50% compared with the AANAP algorithm;the splicing speed is better than the SPHP algorithm and the AANAP algorithm.Compared with the existing large parallax image stitching algorithm,the algorithm stitching result can restore the real scene.The paper has 32 pictures,4 tables,and 52 references.