The Research On Image Stitching Based On Feature Point Matching

Image Stitching is an important research area of computer vision, image processing and computer graphics all the time. Large field of view and high-resolution image can used to construct the application is in virtual reality, medical image processing, remote sensing and military affairs.The paper summarizes the effort of image Stitching that the researchers have achieved. The thesis introduces image the steps and classification of Image Stitching.Chiefly, this paper did research on image registration and mosaicing technique and presented a fast and robust algorithm based on feature point matching.Based on the analysis of the exiting methods and the researches of wavelet pyramidal matching algorithm, image pyramid is used to solve the problem of the high cost of the existing method. In this method, image pyramid is constructed by the wavelet transformation, the images in coarse layers are matched based on feature points, the matching starts from the coarsest layer and the result is up-sampled as initial estimation for the finer layer, after getting initial matching points on the original image, use an RANSAC algorithm to estimate transformation matrix between two images; the transformation matrix is used to guide local point matching again, then combine RANSAC algorithm and Levenberg-Marquardt optimization to decrease registration error. The experiment shows better results than existing methods on matching speed and the robustness of the algorithm.Then, the proposed algorithm is applied in the image stitching, and analyze the disadvantages and advantages of several image mosaicing technique by the experiment.Lastly, the lack of the research in the paper is summarized, then look forward to the futurity of the development of image Stitching.