Research On Image Stitching Technology Based On Homographies

The image splicing technique is to splice two or more images with overlapping areas into a wide-view and high-resolution image through a series of techniques.As a research hotspot of digital image processing,image splicing technology has been widely used in three-dimensional panorama,vehicle security,virtual reality technology and satellite remote sensing technology.After deeply researching and analyzing the correlation splicing algorithm,this paper focuses on the splicing algorithm based on multi-homogeneous matrix and the SIFT feature point extraction algorithm.The main work is as follows:Firstly,aiming at the problem of uneven distribution of feature points in SIFT algorithm,this paper introduces the method of uniform detection of extreme points,expands the neighborhood range of extreme point detection,and uses the radius of adaptive non-maximum suppression to determine the neighborhood of extreme point detection.Range,according to the texture of the image automatically determine the radius,so that the final distribution of SIFT feature points evenly.On this basis,the 128-dimensional descriptor generated by SIFT is used to calculate complex,poor real-time performance,and low matching efficiency.This paper designs a unique SIFT feature descriptor construction method.Selecting the circle as the descriptor of the key point of the domain structure,combining the gradient histogram and the gray difference information as the descriptor's characteristic information,finally forms a new 36-dimensional feature vector.The feature points of the improved SIFT algorithm are more evenly distributed,which reduces the complexity of the algorithm to a certain extent and improves the matching efficiency and accuracy.Secondly,in the image registration phase,the traditional algorithm can not solve the problem that the image has multiple planes.In this paper,the method of multiple homography is used to divide the image to be stitched into multiple squares.Then the homography matrix of the image in the corresponding square is solved by the projection transformation so that the image regions in the corresponding square can be aligned and aligned.This overlapping area can be precisely aligned and has minimal local distortion.After the projection transformation of each local area is completed,a global similarity transformation is used to constrain the projection transformation of each image,making it similar to the similarity transformation as a whole,which can effectively reduce the distortion of the image.