| Image stitching is a process of registering and fusing multiple image sequences with certain overlapping regions and small viewing angles to form a wide-angle image with more information.Among them,panoramic image stitching as a main application direction of image stitching technology,has become a research hotspot in the field of computer vision and digital image processing.This paper mainly studies the basic principles and key technologies of feature-based panoramic image stitching.Firstly,aiming at the problems of high complexity and low registration efficiency of traditional SIFT algorithm,a fast SIFT feature detection algorithm is proposed based on the overlapping region,which greatly reduces the computation time.Then,in the process of SIFT feature matching,the process of transform matrix parameter estimation is simplified by dividing the overlapping area into blocks,which reduces the number of iterations of feature matching.At the same time,by verifying the correctness of the transformation matrix,the correctness of subsequent feature point matching is ensured,and the matching efficiency is improved.Secondly,the stitching of dynamic scene images can not solve the exposure difference and fusion ghosting.Based on the idea of dynamic programming,this paper proposes an improved algorithm that expands the optimal seam search criteria combined with texture differences,which combined with a piecewise linear weighting to eliminate exposure differences and achieve ghost-free fusion.Finally,based on the improved algorithm proposed in this paper,the cylinder projection model is used to realize the cylindrical panoramic image stitching.Experiments show that this method not only eliminates ghosting,reduces the impact of exposure differences,but also improves computational efficiency. |
