The Research Of Image Regiistration Technology Based On Feature Points

Image registration is the key technology of image processing. Image registration research theory has achieved a lot of important achievements. And in many fields has successful appl ication. However, image registration in the theory and the application problems still exist need to address the key and difficult problem. At the same time as the depth of the application of image registration adaptability, registration precision and computational efficiency put forward higher request.According to these needs, this paper studies the feature points based on image registration technology, the purpose is to solve a complex space transformation between existing image of the fast and accurate registration problem. Provide the new algorithm and technology for the solution of the image registration.The research content includes the fol lowing four all aspects.1. The general process of image registration, especially the coordinate transformation and interpolationThrough to the diversified registration methods, present regi strati on method of basi c steps. And in thi s part, introduce and discuss the coordinate transform and interpolation method2. Analyse and improve the feature point detection methods.Carried out experiments for Marvoc algorithm, SUSAN algorithm and Harris algorithm use some images.Analyse the main properties of the algorithms. Presents an improved Harris corner detection algorithm.In the new algorithm, to feature point detection using the multi-scale corner detection technology. The algorithm use a new response function, improve the algorithm of noi se performance. Screeni ng pixel before test, reduce the operating time.3. The study of feature point matching algorithms.This paper analyse the traditional triangle matching feature points method,2 d clustering method, descriptors matching method, point set matching. Analyse the experimental re-sults. Improved the algorithm through fusing informed research. The experimental results show that, compared with the existing algorithm, the new algorithm has better effect.4. Solve the image transformation parameters.In the new feature points matching algorithm,following the improvement of matching points algorithm, Put forward the improvement of solution to transform parameter. Analyse the efficiency of solving method.Although no example proving the method, the new method is more convenient and efficient form the view of efficiency analysisHowever, this research learning work also has unaccomplished matter. Specifically in the following aspects:1. The types of tested images are not comprehensive.The pictures using in this paper include self-time photos and download images, no use aerial images and medical images. I will use classic picture in the later experiments.2. The problem of solving transform parameter.In this paper, main content i s feature point detection and point matching algorithm.Point matching algorithm involved transform parameter, but we didn't make further research ineto transform parameter. The later work still need to carry out in this respect. 3. The procedure realization of the image matchingExperiment didn't realize image matching results. I will do image matching program realization in the later work.