| With the popularity of handheld image acquisition devices such as smart phones and digital cameras,the demand for high-information images is increasing day by day.In the field of optical imaging,high information and high resolution are always the factors that restrict each other.Image stitching technology can expand the field of vision and realize the information integration of the image while maintaining the high resolution of the image.Image registration is the most important step in the process of image stitching.Based on the SIFT feature extraction algorithm,this paper makes an in-depth study of image registration and stitching under different conditions from two aspects:improving the feature extraction algorithm and improving the stitching accuracy.The specific work is as follows:1.In order to improve the efficiency of SIFT algorithm and reduce the matching complexity,an improved SIFT binary feature description matching algorithm is proposed.After the scale space is constructed by SIFT and the feature points are determined,the first-order central moment is used to accurately estimate the principal direction of the feature points;then the binary feature descriptor is determined by using the internal and external magnitude relationship of the gradient in the sector neighborhood of the feature points;finally,the hamming distance between the feature vectors is calculated by using the LSH sensitive hash function,the similarity between the feature point descriptors is evaluated,and the feature point matching is completed.On the Mikolajczyk dataset,the proposed algorithm can not only ensure the registration accuracy,but also improve the efficiency of the algorithm.2.1n order to improve the accuracy of detail region registration and improve the stitching effect,an image stitching optimization algorithm based on image two-dimensional information entropy to distinguish residual weight is proposed.Firstly,the SIFT features of the image are extracted,and the RANSAC algorithm is used to exclude the outer points to get a consistent set;then the image feature points are divided into efficient points and inefficient points by using two-dimensional information entropy,and the corresponding weights are given to the feature points on the consistent set of inner points;finally,the Levenberg-Marquardt algorithm is used to minimize the residual sum,calculate the optimized homography matrix,and then the global projection is used to get the high-precision mosaic image.On the Garden dataset,the proposed algorithm can achieve high-precision image stitching.3.In order to improve the problems of ghost and dislocation in stitching large parallax images,an image stitching algorithm based on feature clustering is proposed.Firstly,the SIFT feature of the image is extracted,and the Tyson polygon is constructed according to the distribution of the matching feature points in the overlapping region of the target image;then the improved AGNES hierarchical clustering is used to cluster the feature points,combined with the stitching error to determine the number of groups,and merge the Tyson polygons in the corresponding group to get multiple image sub-planes;finally,the homography matrix of each region of the image is calculated,and the local projection is carried out to get a complete mosaic image.The proposed algorithm can achieve high-precision stitching of parallax images on Railtracks and Temple datasets.Figure[26]Table[6]Reference[65]... |
