Research On Parallax Image Stitching Based On Local Feature Matching

Image stitching is one of the important branches of image processing technology,which can stitch multiple images with overlapping areas taken in the same scene into a panoramic image with a wider perspective,and is widely used in virtual reality,medical imaging,map mapping,automatic driving and other fields,and has greatly improved people’s quality of life.However,the stitched image obtained when the viewpoint position changes still has problems such as local distortion,ghosting and stitching seams,etc.How to reduce or even eliminate the effect of parallax to generate a visually good stitched image is the core problem to be solved in this paper.In this paper,the image registration and image fusion algorithms in the image stitching process are studied in depth,and the parallax images are stitched using the designed registration algorithm and fusion algorithm,which effectively eliminates the problems of image ghosting,distortion and stitching seams caused by parallax,and at the same time,an image stitching system is designed to provide users with a better interactive experience.The main work and innovations in this paper are as follows:1.A feature optimization-based image registration algorithm is designedAiming at the problems of the current feature matching algorithm such as insufficient number of feature matching and unsatisfactory registration effect due to excessive time consumption,this paper designs a feature optimization based image registration algorithm.Firstly,a circular region is used to replace the rectangular region in the traditional SIFT algorithm to generate a 60-dimensional feature descriptor,which reduces the computational complexity and improves the operation speed.Then,a novel feature matching strategy is used to filter the interior points to improve the robustness of the image registration and better eliminate the mis-matched points,and at the same time,linear features are introduced and dotted to enrich the feature point types.Finally,the extracted feature point pairs are used to calculate the mapping relationship to align the actual positions of the images.The experimental results show that the algorithm of this paper effectively solves the problems such as insufficient number of feature matching and long time consuming.2.A weighted fusion algorithm based on circular function is designedIn the actual image acquisition process,differences in equipment parameters,lighting changes and the influence of the surrounding environment can cause different degrees of parallax and luminance differences in the images,in addition,the image registration will produce certain errors,making the aligned images appear ghosting and obvious stitching seams.To address the above problems,this paper designs an image fusion algorithm based on the weighted circular function,which is based on the idea of fading-in and fading-out fusion algorithm,using the circular function to weight the overlapping areas to effectively solve the problem of unnatural image transition.The experimental results show that the algorithm in this paper effectively suppresses the local ghosting problem and eliminates the stitching seam phenomenon at the same time.3.An image stitching system is designedThe system contains functions such as system initialization,image display,image registration and image fusion algorithm selection,etc.The algorithm designed in this paper is used for image stitching,and direct average fusion algorithm and linear weighted fusion algorithm and the circular function weighting based fusion algorithm designed in this paper are also provided for users to choose,so that users can operate the complex steps of image stitching simply through a visual interface,which greatly enriches the user experience.