Image Registration And Application Based On Feature-point Detection

Image registration, which is not only a fundamental problem but also a precondition and key one, is a process to match two or more images of the same scene taken at different times, from different viewpoint, or by different sensors. The methods of image registration can be classified into two parts: intensity-based matching approach and the feature-based matching approach. In this paper, the theories of two categories are dissertated. Feature-based point, which is considered as one of the most popular method by now, has the largest advantage on its ability of translating the analysis of the whole image into its feature points. In that way, it speeds up the image process. So the feature-based points matching approach is discussed as a main point.The exactness of feature detection and orientation has great influence on the matching process. In my paper, the methods of feature point detection are discussed, and the methods of feature points detection and matching based on Harris operator and Susan operator are discussed in detail and realized by experiments basing on the criterion of NCC. Then the image matching method based on SIFT operator is emphasized. And by analyzing the matching qualifications of SIFT features, the extending matching method based on the SIFT feature points is put forward by using the method of searching feature points in both two ways. This new method reduces wrong matched points and increase the precision. It has been proved effectively by the experiments.Finally, an important application field, image mosaic, is put forward. Based on registration method in this paper, a new image fusion method is discussed and a new platform of image mosaic has been constructed. By using the technical of supplement with color, we solve the problem of chink during mosaic. The comparing experimental results prove that the method can significantly reduce the time of matching and solve the problem of joint.