Study Of Registration Algorithm In Image Mosaic Based On Key Points

Image mosaic is an important branch in image processing field, and it is widely used in the fields of computer vision, image-based rendering technique, medical image processing and 3D reconstruction.This paper focuses on the method of image registration based on image key points. Generally, image mosaic method based on key points includes features detection, matching based on interesting points, and image blending.For key points detection, Harris corner detector and SIFT descriptor is adopted in the paper because they are robust to noise affection compared with other point features. NCC method is adopted for the matching of Harris corners and Euclidean distance method is adopted for SIFT descriptor in the initial matching of the image key points.A new image registration method is presented by the author in this paper for image registration. In this method, there is a new twice guided-matching scheme based on RANSAC algorithm. The transformation of the images in this paper is supposed to be affine transform, firstly the inliers set is got through RANSAC algorithm, then the registration error threshold is calculated by RANSAC algorithm again, finally the affine matrix is computed by comparing the registration error with the 90% value of the threshold got in former step. The affine matrix is computed and verified by 6 points in this paper, and this is improved comparing with the existing algorithms and which makes the result more accurate and the affine matrix more satisfied. The convergence of this method is better than LM method.Several typical blending methods are given in Chapter Five.For key point detection and image blending, the paper provides some examples to test the effect of image mosaic, and the author also compares the different methods.