Research On Image Stitching Algorithm Based On Feature Optimization And Natural Warp Fusion

Image stitching is a technique to warp and fuse images with overlapping areas of the same scene taken from different viewpoints into a single panoramic image with a larger perspective,which is widely used in computer vision,pattern recognition,medical imaging,camera panorama and remote sensing.Due to realistic factors such as parallax,low texture and wide baseline,existing image stitching algorithms have poor robustness,and image artifacts,pixel blurring and object structure distortion occur in stitched images.In order to solve the above problems,this paper carries out the research of image stitching algorithm based on feature optimization and natural warp fusion,and the main work is as follows:First,local image pre-warp for Bayesian probabilistic model and global image prewarp for projection invariant feature number are proposed for image alignment and alignment problems.Based on the characteristics of local image warp model,the feature interior point optimization of Bayesian probabilistic model based on grid weight decomposition is designed to avoid the deficiency of solving multiple models with excessive errors in RANSAC algorithm.Projection invariant characteristic number feature interior point dataset optimization improves image alignment by constructing geometric structures to add feature anchors,while reducing the parameter estimation of the model under the global warp of the camera model and improving the robustness of image alignment.Secondly,the fusion of the similar transformation model and the fusion of the global line grid warp model are designed for the two proposed pre-warp models to address the distortion problem arising from the pre-warp of images.The fusion of similar transformation models is designed in the Bayesian probabilistic model of local image pre-warp model to suppress the overall image distortion while ensuring image alignment,and to set rationalized parameter ratios to prevent over-similar transformation.A global line grid warp model is constructed under global image pre-warp of projection invariant characteristic number,and the image structure is protected by designing a global line merging algorithm and a global line term to suppress the projection distortion and perspective distortion of the image using a minimization energy function constraint term.Finally,linear fusion stitching and image-aware seam stitching are used to complete image stitching for the two different warped fused images to be stitched,respectively.In local warp fusion,linear fusion is used in the overlapping region using its whole image division grid warp property,and image pixel weighting is assigned to make the image stitching overlapping region smoother.Image aware seam estimation is used in global warp fusion to find the best stitching seam of an image by chromatic aberration perception and object detection to achieve image stitching.Finally,subjective visual assessment and objective quantitative evaluation were conducted by selecting different image datasets to verify the effectiveness of the algorithm in this paper.