Research Of Image Mosaic Based On Image Registration

Computer graphics is a subject of showing, processing and displaying the graphics. And it is one of the most universal branch of the computer science. With the development of computer technology, image based rendering (image based rendering, IBR) technology which draws a new image from original images have got an intensive study.Image stitching technology is an important area of study in image-based rendering (IBR). It makes a set of overlapping image which (is got from) the same sensor shoot at the different time or the different sensors shoot at the same time to stitch a big image with high definition and wide viewing angle. Image mosaic technology has been widely used in virtual reality, medical treatment, remote sensing technology, aerospace and so on.This paper makes a lot of research on the whole process of image stitching technology, and proposed a approach of image registration which is based on improved K-means clustering and simulated annealing algorithm, which is a novel idea different from traversing all the feature points in the past. In this paper, Harris corner detection, Canny edge detection and SIFT feature extraction algorithms are introduced, and the experiment results are analyzed. This paper presents an approach of cluster with the initial centers obtained by the density function, which avoids uncertainty of the results of the clustering because of the arbitrary selection of the initial center. Optimize the results of clustering which is used as initial results of simulated annealing algorithm by perturbation. Finally, smooth images by calculating the homography between images, and apply fade-in and fade-out to stitching images. Experimental results prove that the method is robust, and can be applied into image mosaic, and also can be applied to the image stitching under the complex environment.