Research On Suppression Methods Of Edging Parallax Artifacts And Geometric Discontinuities In Image Stitching

The research and application of image stitching technology have made great progress.However,in the existing parallax image stitching algorithms,there are still two problems of edging parallax artifacts and geometric discontinuities,which seriously affect the quality of parallax image stitching.Therefore,the researches on edging parallax artifacts and geometric discontinuities suppression methods in image stitching have important value and significance.The problem of edging parallax artifacts appears in the parallax image stitching algorithm with mesh-based alignment optimization,especially when there are obvious changes in the edge depth of the object in the image.How to give the feature points on objects with different depths and the local homography calculation weights with obvious differences is the difficulty in suppressing edging parallax artifacts.The problem of geometric discontinuities appears in the parallax image stitching algorithm based on seambased optimization.For complex scene images with multiple objects,the seamline may cross various objects on the ground.How to construct a more reasonable energy function to ensure that the seamline only crosses a single object is the difficulty in suppressing geometric discontinuities.Aiming at the edging parallax artifacts problem existing in the current mesh-based alignment optimization algorithm,this thesis uses the warping residual vector to realize the quantitative description of the feature point depth information,and integrates it into the weight matrix calculation formula of the mesh-based algorithm to ensure the feature points on different objects have obvious weights.In the stitching comparison experiment on the public image data set,compared with the sub-optimal algorithm,the algorithm in this thesis reduces the average value of the image alignment evaluation index MSE by 33.18%.In the real image stitching comparison experiment,the algorithm in this thesis also effectively suppresses the parallax artifacts on the edges of objects.Aiming at the geometric discontinuities problem existing in the current seam-based optimization stitching algorithm,this thesis uses the pixel difference of the corresponding pixels caused by the image parallax to realize the quantitative description of the depth information of the image overlapping area,and integrates it into the construction of the energy function to overcome the defect that the traditional energy function cannot guarantee seamline cross only a single object,and effectively suppresses geometric discontinuities.The stitching comparison experiment results on the image data set constructed in this thesis show that the algorithm effectively suppresses the geometric discontinuities that may occur in the seam-based optimization algorithm.In 40 sets of stitching experiments,the average structure damage degree of the algorithm proposed in this thesis is 36.70% lower than that of the sub-optimal algorithm.