Research On Image Matching Algorithm Based On Filtering And Dynamic Sampling

Image matching technology is an important technology in the field of computer vision,and has a wide range of applications in many fields such as image stitching,3D reconstruction,and SLAM.However,due to the influence of external noise factors such as light,occlusion,viewpoint transformation and deformation,the matching results obtained by various image matching algorithms have certain limitations,so it is a valuable research to choose the appropriate algorithm to get the exact matching of images.In this paper,we propose an image matching algorithm based on filtering and dynamic sampling for images with long baselines or large viewpoint transformations,which is divided into two parts(1)a filtering model based on the geometric topological distribution characteristics of matching points: this model uses the distribution characteristics of matching point pairs to construct a filtering model to remove a large number of incorrect matching point pairs in advance,effectively improve the probability of correct matching point pairs,pave the way for the subsequent work,reduce the number of iterations of the algorithm,and reduce the algorithm running time;(2)dynamic sampling algorithm based on Markov Chain Monte Carlo(MCMC): this method gives each matching point pair a certain probability,making it dynamic,and changing from passive to active in the previous algorithm.It can speed up the finding speed of the correct model.In the whole algorithm design process,the algorithm in this paper proposes a data filtering model based on the geometric topological distribution of matching points according to the geometric topological distribution of the interior points;it fully considers the randomness of matching points and proposes a sampling consistency algorithm based on Markov Chain Monte Carlo,which effectively solves the problem that images with long baselines or large perspective transformations will have a large number of wrong matches in the matching process.The results of this paper show that(1)when there is no obvious scaling between images,even if the viewpoint difference is small,the geometric distribution of correct matching point pairs(inner points)between images is closer,and the difference of distances between inner points is less than the difference of distances between wrong matching point pairs(outer points)with high probability;when there is obvious scaling between images,the difference of distribution between matching point pairs between images is When there is significant scaling between images,the difference in the distribution of matched point pairs between images is more obvious,and the mean value of the distance between inner points is smaller than the distance between outer points.The filtering model thus constructed can effectively remove the outer points and improve the probability of inner points.(2)Since the matching point pairs are given vitality,it makes the sampling more efficient and greatly reduces the time to find the correct model.(3)After comparing with other algorithms,it also proves that the proposed algorithm has better advantages in targeting the problem that a large number of false matches occur in the matching process for images with long baselines or large perspective transformations.