The Research Of Image Matching Technology Based On Feature Detection

Image matching is the implementation process of image registration technology. As the basic problem in image processing, this technology is the key step of many image analysis and processing tasks, and has a very important research significance and practical value. At present, the image matching method can generally be divided into two categories: region-based image matching method and feature-based image matching method, region-based matching methods can also be divided into matching method which based on gray scale or transform domain. After comparative research and analysis in various types of image registration method, the paper focuses on the feature-based image matching algorithm. Among them, image matching method which based on feature points has less calculation, more accurate positioning and stronger robust to image transform, light change and noise. The traditional image matching methods based on feature extraction, such as Harris algorithm, SIFT algorithm, have drawback like large computation cost, low efficiency and so on. This paper presents a method which uses image edge curvature feature points for image matching.Image edge is the most important invariant feature, which has stable property, less susceptible to noise and light transform effect et... Compared with the traditional edge curvature computing method, this paper uses polygon approximation simplification edge information, through the Freeman code to extract the approximate curvature of the polygon, and the extreme points are the feature points for image maching. Then build up the characteristic descriptors for feature points. Meanwhile, this paper analyzed the characteristics of the typical process of matching similarity measure algorithm, based on the traditional Euclidean distance, weight set method and Hausdorff distance algorithm, put forward improved image matching algorithm which based on the characteristics of feature points subset of the Hausdorff distance and the triangle areas computed by neighborhood feature points.The experimental results show that the image matching algorithm based on edge curvature feature points does not lose performance in terms of robustness and accuracy compared with traditional algorithm. At the same time, the algorithm also has good images matching results with noises, illumination change and different scale-space.