Research And Desigh Of Image Matching Algorithm Based On Local Features

Computer vision technology which is an emerging field has attracted more and more attention. Image matching is the key technology in computer vision, and its prospects are very bright. Medical imaging, remote sensing, industrial product testing and so on are inextricably linked with the image matching technology. In this paper, against image matching technology based on the local features the following work is mainly done.(1) There are more false match points against local features in the process of landmarks matching, which affects the reliability problems matching results. Local Feature and RANSAC algorithm combined solutions is designed. Firstly, the traditional algorithm of SIFT and SURF is studied, and then the similarities and differences between the two are compared and analyzed. Secondly, simulation experiment is done by using SURF algorithm on landmarks matching, Experimental data shows that there are more false match points in the matching process which have a great impact on the reliability of the results of the match. Finally using RANSAC algorithm effectively eliminates a lot of false match points to improve the reliability of the matching results, mismatched points which are not eliminated are analyzed to point out the shortcomings of RANSAC algorithm.(2) It is inefficient against the problem of specific figures match by using the single local features, we design a solution based on the weight of the combination of local features. Firstly, simulation experiment is done against the single local features in particular face matching, the experimental results show that the matching is inefficient, so we analyze several important factors that affect the results. Then combining the important factors that affect the results,in the establishment of personal sample library, extracting feature points in the selected M representative images form a pool of features. Geometric restrictions is introduced during the processing of a single sample image, and in the region of geometric restrictions we extract feature points according to different parts of different weights. Experimental data shows that the improved algorithm increases the matching accuracy.(3) From the requirements analysis of the project,we design the overall architecture and function module. And analyze the algorithm of the specific characters in the system, to explore its feasibility.