Feature Points Matching Algorithm Research Based On Multiplex Method

Image matching is a basic problem in the field of image processing. It includes gray-based method and feature-based method. The former method is succinct and has wide application, but the time complexity of the algorithm is high, especially it's difficult to deal with the situation while the image has rotation and scaling. The latter method is much more easily overcoming the difficulties which are encountered by the former method, but how to create the corresponding relationships between the images' features is always a difficult problem. A new feature point matching method is proposed in this thesis, which is based on similar triangles approach, together with 2-D parameter cluster approach and descriptor-based approach.The matching method based on feature points contains two steps. They are as follows: extracting feature points and matching the feature points. Extracting feature points, as the first step of the method, can directly impact the matching result. After analyzing and comparing some of traditional methods of extracting feature points, this thesis selects the Harris corner detection for extracting feature points from gray images because of the well experimental result. However, the traditional Harris corner detection is not suitable for the situation while the image has scaling. The improved Harris algorithm having the function of anti-scale is implemented by using scale space in this thesis.After analysis and comparison, the traditional method based on similar triangles is easy to be implemented and has good robustness, but it requires judging the co-rotating similarity between each couple of triangles, so the time complexity of this algorithm is high, and the stability of this algorithm depends to a great extent on the extraction of feature points. 2-D parameter cluster approach, as another feature-based method, has the advantage of high efficiency, but this method has high requirements to the number of the initial effective points. Descriptor-based method can process the images which have certain deformation and perspective transformation, but this method is complicated to be implemented and has high requirements to image texture. Then in this thesis, a fusion feature-based matching method is proposed, which is based on traditional similar triangles approach, 2-D parameter cluster approach and descriptor-based approach. It greatly reduces the request of the initial effective points.The time complexity of the algorithm is decreased form 0(n~3) to 0(n~2) by using active triangles retrieval method to replace the traditional passive retrieval method in the thesis. The efficiency of the algorithm is greatly enhanced by realizing the improved algorithm and parameter cluster in vector space. Meanwhile, the matching results become more precise by adopting a dynamic minimum-distance clustering.At last, the matching method in this thesis is applied to fingerprint identification while the images of fingerprint have translation, rotation and scaling. The result shows that the method is stable, rapid and accuracy.