Research Of Image Registration And Stitching Based On PCA-SIFT

With the continuous development of the information society, data, video, image, text,etc. are used in all walks of life. Among them, image is extensively used. Thus, the image processing technology has become a hot topic. However, wide viewing angle of the image used more often in the application of research and projects in various fields. But,general camera view is small, you can not get a wider field of vision. Therefore, image splicing processing is needed.Before image stitching, the most important task is to conduct image registration.Currently, based on PCA-SIFT image registration has slowly become a hot spot.However, the commonly feature extraction and characterization of consume used more time, and the matching accuracy in general. Improvement of the feature descriptor aiming at improving the efficiency of the registration is also developing, at the same time,ways of image stitching to eliminate the gap is gradually increasing. But the mosaic effect on color images is not ideal.This paper studies the basic theory of image registration, describes the basic theory of SIFT and PCA-SIFT method, focusing on describing the shortcomings of traditional SIFT feature point descriptor, and do simulation experiment with Matlab, compared the accuracy and time efficiency of image registration process, analysis its strengths and weaknesses.Previous matching algorithm need more storage space and time. In this paper, SIFT algorithm is improved, and the method of weighted variable ring is used to describe feature points, and PCA dimensionality reduction techniques is used to reduce the dimension from 128 to 20. the experiment validates the improved method in time complexity.Finally, In view of the present most of the algorithms in the process of image Mosaic still have obvious crack defects. This paper presents the best point smoothing method: Firstly, extract image feature points by the method of weighted variable ring,and reduce the dimension of feature point descriptor with PCA-SIFT; Secondly,eliminate the false match point by RANSAC, use the color closest points as the splice points; Thirdly, in order to obtain the best splicing line, splicing point set is need for further smoothing. Simulation results show that the proposed algorithm can get seamless image while maintaining the image of the original information, and has improved on the efficiency and effectiveness.