A Method to Eliminate Wrongly Matched Points for Image Matching
In this paper,we proposed a new method to eliminate wrong point pairs by distance constraint and bidirectional matching.It was aimed at the problems of many wrong pairs and poor matching accuracy in feature-based image matching.It calculated the distance of point pairs obtained by SURF,eliminated part of wrong point pairs by this distance,and then got the best point pairs by bidirectional matching.Used this point pairs,it calculated the homography to mark out the boundaries of reference image.At last single image and video were used to verify the validity.The results indicated that the method can eliminate most wrong matches.The accuracy was improved by 3.07%compared with RANSAC,the speed of single image matching was 0.23s,the frame rate of video matching reached 25.108fps,and the matching speed of SURF was also guaranteed.
Xiao-fei AI
Hangzhou Dianzi University,China
国际会议
上海
英文
9-13
2017-12-30（万方平台首次上网日期，不代表论文的发表时间）