The Algorithms Of Ghosting Elimination And Stitching Line Fusion In Image Mosaic

With the rapid development of computer vision,it is more and more demanding for the video and image,including the large scale and wide angle panoramic image.Image mosaic technology combines the collected image sequences into a seamless and high definition panorama to meet pepole's visual requirements of video and image.Because its application prospect is very wide,many scholars have conducted in-depth study of the image mosaic algorithms.Image mosaic technology is divided into two types: one based on feature and the other based on region.Image mosaic based on feature has the advantages of small computation and high real-time.It is a hot spot in the research field.Firstly,in order to improve the real-time performance of image mosaic,produce the seamless panoramic mosaic and improve the quality of image mosaic,the technology based on feature was researched in this thesis.Secondly,the traditional mosaic algorithm was improved,and a seamless image mosaic algorithm based on improved registration and fusion was proposed to boost the real-time and the quality of image mosaic.At the last,a fast seamless image mosaic system was designed and implemented.The effectiveness and practicability of the system was proved by experiments.The research work of this thesis includes:1.In the process of constructing the Pyramid image,the traditional SIFT feature point detection algorithm used the linear Gaussian expansion filter to smooth inner area effectively.However,the Gaussian filter sacrifices the local accuracy and it was easy to cause the boundary information loss.Besides this,a large amount of time was used to extract the feature points.In order to increase the accuracy of positioning the feature points and shorten their extraction time,a feature point extraction algorithm based on nonlinear filtering(A-KAZE)was used in this thesis.Then,traditional KNN matching algorithm was discarded and a two way matching model was adopted based on it.This algorithm improved the saliency and precision of feature points' registration.2.For the ghosts caused by moving objects in the image,and the problem of uneven color transition at the seam,the traditional fusion algorithm was improved and the Laplacine fusion algorithm based on stitching line was proposed.By limiting the image fusion in the range around the stitching line,it guaranteed the fusion efficiency,shortened the fusion time,eliminated the seams and ghosts effectively,and increased the quality of panoramic mosaic results.3.The camera rotation may cause the rotation of the world coordinates,which leads to panorama tilt and distortion.At the same time,the large number of accumulated errors will result in panoramic distortion when multiple images are stitched.A method of automatic panorama straightening was presented to solve these problems.4.In this thesis,a fast seamless image mosaic system was designed and implemented to eliminate the ghosts in the process of stitching,which could realize the seamless mosaic for the input image sequence and achieve high quality of panoramic image.