Research On Matching Algorithm Based On Image Feature Extraction And Description

Image matching is one of the most important problems in the field of image processing.It refers to finding two or more images in different shooting environments and finding one or more transformations in the image space,and establishing the mapping relationship between images,so as to complete the matching process.At present,it is used in the fields of visual navigation,target tracking,image mosaic and so on.But under the influence of the shooting environment and the defects of sensors themselves,how to improve the matching rate,robustness and timeliness of the image matching algorithm has become the focus of the research.Feature based image matching algorithm extracts relatively stable image feature information from image to match feature points.Compared with it,it has the advantages of high efficiency and robustness.Therefore,this paper mainly studies the image matching algorithm based on feature matching algorithm,through a variety of classic,especially in the study of FREAK algorithm and BRISK algorithm,the image feature points of improved FREAK and RANSAC algorithm,and the edge of the image and the image feature points matching algorithm based on improved BRISK.The main contents of this article are as follows:(1)Systematically introduce the main process of feature-based image matching algorithm: the extraction of feature points,the description of feature points and the method of matching points to purify,and introduce some commonly used algorithms in each process.(2)An algorithm of image feature point matching based on improved FREAK algorithm is proposed.The algorithm firstly generates Gauss differential Pyramid image,the feature points detected with scale invariance;then the feature points on the description of FREAK descriptors,obtain binary descriptors;finally,through the Hamming distance matching coarse matching feature points in the process of matching,and the matching points of thepurified RANSAC algorithm,realize the feature two image matching.Experimental results show that the improved algorithm proposed in this paper effectively solves the problem that FREAK does not have scale invariance.When the scale of image changes,the matching accuracy of the improved algorithm is higher than that of SIFT and FREAK algorithm.(3)An image feature point matching algorithm based on edge image and improved BRISK is proposed.First,the edge detection of two images by Sobel operator;then based on the SURF algorithm is adopted to extract feature points of the two edge images;then for each feature point BRISK algorithm is used to describe the formation of binary feature vector descriptor;finally,in the feature matching stage using Hamming distance of two-way matching points combined with the RANSAC algorithm to remove the error,to complete the two images to match the feature points.Experimental results show that the improved algorithm effectively solves the defect of uneven distribution of feature points caused by fixed threshold value,which shows better robustness and correct matching rate than BRISK algorithm.