Image Stitching Research For Complex Moving Foreground And Large Parallax Challenges

As an important research in computer vision,image stitching aims to fuse a group of images with overlapping areas into a wide-view panorama.Although image stitching has received extensive progress,it is still challenging to obtain a more natural panorama.Based on the mainstream stitching scheme,this paper proposes an image stitching research for complex moving foreground and large parallax problems from the perspective of practicality.Specifically:To address the problem of matching difficulties and eliminating ghosts,an image stitching algorithm via foreground Multi-Matching and stable structure preservation is proposed.Firstly,the dynamic objects are extracted the deep features and then established matching relationship based on Multi-Matching,so as to solve the matching difficulties.Secondly,the energy terms will be optimized,which mainly include alignment item,line preservation item and distortion control item.This module reduces parallax ghosts and distortion by ensuring high-quality alignment and enhancing structural fidelity.Compared with mainstream image stitching frameworks,the extensive experiments demonstrate that the proposed method can provide more natural stitching results,and achieve better performance in PSNR and SSIM.To address interference on feature detection and matching in pre-alignment stage,global linear structure stretching and distortion caused by complex foreground and large parallax,an image stitching algorithm via feature augmentation selection and homologous model optimization.Initially,the selection strategy is implemented by using Earth Mover’s Distance to improve SIFT similarity metric and setting weighted factors,which avoid the feature mismatching problems.Next,a global line generation strategy is designed to optimize the homologous model by using the collaborative constraints of local lines,global lines and adjacent meshes,which effectively avoids the stretching and distortion of linear structures.Experiments show that the proposed algorithm performs better,effectively improves the accuracy of stitching,and the RMSE value is improved by about 37%.