Research On Image Keypoint Extraction And The Application On Registration

Corner extraction is one of the topic in the field of image processing and computer vision. Corner extraction has the unshakable status in the fields of object recognition, image classification, image restoration, camera autofocus, target tracking, stereo matching, robotic navigation, and video, and so on. The basic process of corner extraction has two steps:First, using corner detection algorithm detect the corners’position, scale and so on; Second, using the information of detected corner and its neighborhood describe the corner. The research of this paper is image corner extraction and the appplication on registration, including the following aspects:1) This paper has made the further study on several typical kinds of feature extraction algorithm.2) According to the characteristics of the technology of image registration, this paper has made further studies on multi-scale analysis and clustering analysis technology; In the further study of the image pyramid, this paper had done a lot of experiments; and studied and have optimized the K-Means algorithm which is one of the most common used clustering algorithm; Eventually, combined with multi-scale analysis and clustering analysis technology this paper has optimized image registration techniques.3) This paper has made further study on the image matching algorithm based on corner extraction and used adaptive window instead of a fixed one to reduce manual intervention, and carried on the simulation and analysis.4) Combined with multi-scale analysis and clustering analysis technology this paper has designed a multi-scale clustering related image registration method.Firstly this paper establish image pyramid, and then choose the appropriate small scale images to extract the feature points and match. We find the class which has the most features by clustering is the overlap area. This paper just extract and match the feature points in the overlap area of the large scale image, and remove the false matching feature points calculating the transform parameters of the registration image by the RANSAC algorithm. The method improved the speed of image registration, and has more accurately transformation parameters of images, so it improved the precision of the registration.