The Algorithm Of Image Mosaic Based On Binary Tree And Eliminating Distortion Error

With the rapid development of computer vision technology,images with narrow viewing angles can not meet the needs of people.Image stitching technology is to register and fuse multiple continuous images in the same scene to synthesize panoramic images with wide viewing angle and high resolution.The application prospect is very wide.How to improve the real-time and stitching quality of image stitching is a hot topic at home and abroad.Image mosaic technology is mainly divided into image mosaic based on gray information and feature.The former is realized by calculating the similarity of two images to be spliced and determining the overlapping region,which has large computation and low robustness.The latter is mainly used to extract image feature information and realize splicing through feature point matching,which is fast and robust.The image mosaic based on feature was studied in this thesis.By improving the traditional image stitching method,the image mosaic algorithm based on binary tree and eliminating distortion error was proposed,which accelerated the efficiency of image stitching and improved the quality of image stitching.The main research work of this thesis includes:1.Aiming at the problem of low image registration efficiency and high execution time,the A-KAZE(Accelerated-KAZE)feature point extraction algorithm based on nonlinear filtering was used in this thesis.At the same time,a block feature point extraction algorithm based on A-KAZE was proposed in this thesis.The estimated overlap region of the two images to be stitched was detected by the A-KAZE algorithm based on nonlinear filtering.Then,the bidirectional KNN algorithm was used to match the feature points,which effectively avoided the registration time of unnecessary regions in the image and improved the real-time and accuracy of image registration.2.Aiming at the problem of distortion of panorama after multi-image splicing,an image mosaic algorithm based on binary tree was proposed in this thesis,which was different from the traditional image mosaic method from left to right.The input image sequence to be stitched was set as the leaf node of the binary tree,and a binary tree was constructed from bottom to top.Finally,the image of the root node of the binary tree obtained by the splicing was a panoramic image.3.In the process of image shooting,due to the rotation of the camera and the cumulative error of image matching,the panorama appears tilted and distorted.In order to improve the quality of image stitching,an automatic straightening algorithm for the panorama was proposed in this thesis.It determined whether it is in the straightening range by calculating the distortion of the panorama.If straightening is required,then the vertex coordinates after the panorama being straightened were estimated,and the perspective transformation matrix before and after the panorama straightening was calculated.Finally,bilinear interpolation was used to straighten the panorama and eliminate the distortion.This thesis designed and implemented a fast image splicing prototype system to ensure the splicing efficiency and improve the quality of the spliced panorama.