A SIFT Stitching Algorithm

Image stitching is one of the most important research areas among computer vision and image processing. It’s a task used to match and register a group of images(which are got by different sensors, from different perspectives or at different times) having spatial overlap into a reference surface, and then use blending method to form a seamless panoramic image, which contains the image sequence information and resolves the problem of high resolution and wide viewing angles. Because the image mosaics technology has solved the contradictory problems(small perspective of imaging equipment) between the field of vision and the resolution. It has many applications, such as remote sensing image processing, medical image processing, digital video compression, motion analysis, virtual reality and so on.First, the present research status, application domain and the broad application prospect of mosaics are introduced in this paper. Later, the basis steps and key techniques of image mosaic are discussed. The common image matching and image blending methods are introducedr and their performance as well as the existing problems is analyzed. Based on the former research, this paper describes all automatic panorama generation method using SIFT(Scale invariant feature transform). Firstly, selecting corner collection to be matched by Shi-Tomasi operator detection, subsequently, doing coarse matching by using cross-correlation matching to get many-to-many matching pairs, and then acquiring one-to-one matching pairs by RANSAC (Random Sample Consensus) fine matching, finally, figuring out transformation matrix with matching pairs to stitch panoramas.The main contribution of this paper is:in this paper, different strategies are adopted to shorten the running time in different stages of the total matching algorithm, and the Shi-Tomasi corner detection, SIFT algorithm and RANSAC algorithm are used to guarantee the calculation accuracy. The proposed algorithm takes full advantage of the gray-level information and corner location information of images. Experiments showed that the algorithm could do accurate fast matching of Chromatic aberration or deformation images and it outperforms traditional point matching algorithms both in accuracy and computing speed. It has a large practical.