Research On Image Matching Method Based On SIFT Operator

As a core technology in areas of 3D-reconstruction, target recognition, image stitching, navigation and other areas, image matching accuracy and reliability are the foundation to pro-mote faster development of these fields. Thus improving image matching accuracy become in-creasingly the focus of attention. Based on the analysis of the current image matching on the ba-sis of the research status and the matching technology in this article, from image matching pre-processing, feature point extraction, feature matching and outliers removed four in-depth study, SIFT feature point extraction method, and use a variety of measures to limit match, combined a variety of constraints with RASNAC method gradually eliminate a gross error by the coarse to fine matching strategy is proposed.The main content of this paper includes the follows:(1)Image preprocessing. Compared denoising method of mean filtering, median filtering and adaptive filtering, enhancement method of histogram equalization and wallis operator sharpening, according to the results of the effect of removing the noise and preserving image edge information, adaptive median filter denoising and wallis operator sharpening to enhance the image preprocessing is choosed to achieve the purpose of eliminating noise, enhance image edge information.(2)Feature point extraction. The principle of Moravec, Forstner, Harris and SIFT extraction algorithm are discussed in detail. This paper choose SIFT operator which has strong robustness and sub-pixel accuracy to extract image feature points according to computing speed and the ex-traction ability.(3)Image matching. Match the SIFT feature vectors. Using two measures to restrict match, the one is compared the ratio of the nearest neighbor and the second nearest neighbor with the threshold, the other one is correlation coefficient. Experiments show that using two measures is more precise.(4)Gross error elimination. Combined with homography matrix and epipolar constraint as double constraint conditions, applied RANSAC algorithm to eliminate the error matching. The paper realizes that the strategy from coarse to fine matching improves the initial matching accu-racy(5)Do three experiments for the coarse-to-fine matching strategy which proposed in this paper. The correct rate of match is all over 85% using strategy combined with a variety of measures and multiple constraints to eliminate error through experimental analysis, providing precise matching points for related work such as image stitching and 3D reconstruction and so on.