Research On Wide Parallax Image Stitching Algorithm Based On Slic Segmentation

Image stitching technology is the process of stitching image sequences with overlapping regions into larger viewing images,which solves the problem of smaller viewing for a single camera.However,when processing the parallax scene image sequences,the stitching results often appear the phenomenon of artifacts,accompanied by distortion.In order to solve the above problems,this paper has carried out research on image stitching algorithm in wide parallax scenes,mainly including the following three aspects:Firstly,aiming at the problem of how to extract the majority and uniformly distributed feature points in the wide parallax scene,different feature point detection and matching algorithms are compared and analyzed,and feature line detection and matching are added on the basis of feature points.Then,combined with mesh deformation,a dual-feature local mesh registration model is constructed for pre-alignment to eliminate the impact of small and medium disparity changes.Finally,the line segment retention of dual features is verified by comparative experiments,and the evaluation index of line segment distortion is given for quantitative analysis.Secondly,aiming at the outward expansion distortion in the non-overlapping regions of the pre-aligned image,the Jacobian linearization of the homography matrix is used to improve it.Then,use the Multi-GS(Guided Sampling for Multi-structure Data)algorithm to classify the scene structure of inliers and fit the registration models of different planes.Then the local registration model and the global similarity transformation of the minimum rotation angle are weighted and combined to achieve a smooth transition from overlapping regions to non-overlapping regions.Finally,the quality evaluation index of image distortion is given,and the effect of image naturalness adjustment is analyzed qualitatively and quantitatively through comparative experiments.Finally,aiming at the problem of projective error of overlapping regions in parallax scene,SLIC(Simple Linear Iterative Cluster)is used for semantic segmentation to ensure the consistency of projective content.Then the local registration model fitted by the MultiGS algorithm is used to obtain the projective error sequence by projecting the feature points,and then the accurate local registration model is fitted by using the feature point set with th same projective sequence in the superpixel block.Finally,the feasibility of the scheme is proved by comparative experiments,the registration accuracy and the degree of image distortion are analyzed qualitatively and quantitatively,and the robustness of the image mosaic algorithm in this paper to the change of parallax is proved.