Research On Image Stitching Technology Based On Pseudo-High Dynamic Range

Image stitching as a basic technology in the field of image processing and pattern recognition,has been widely used in computer vision,aerial remote sensing,image registration,reconstructing high-resolution images and medical image processing.With the development of science and technology,people have put forward higher requirements for the accuracy,efficiency and universality of image stitching.This paper focuses on the three core links of image stitching and proposes corresponding innovative algorithms.The main research contents of the thesis include the following aspects:(1)In-depth study and discussion of image stitching technology and basic theory,expounding the overall algorithm flow of image stitching.Through the exploration and comparison of common image stitching algorithms,the image registration algorithm based on Speeded Up Robust Features(SURF)is proposed for image stitching,and the image fusion is performed by weighted average method.(2)Aiming at the problem of low feature matching accuracy in feature-based image stitching in the extreme illumination scenario,an image enhancement algorithm based on Pseudo High Dynamic Range(P-HDR)is proposed as an image stitching preprocessing method.The algorithm generates multiple virtual illumination quantities according to the virtual exposure value and the proportional function,and further multiplies the multiple virtual illumination quantity by the reflection quantity to generate multiple virtual brightness images.Finally,the enhanced image is obtained by performing tone mapping on the multiple virtual brightness images.The proposed image enhancement algorithm can significantly enhance the edges,corners and contrast of the image,effectively suppress image noise,detect more feature points,and improve image feature matching accuracy.(3)Aiming at the problem of the lack of coarse matching algorithm in image registration and the number of feature points obtained by Random Sample Consensus(RANSAC)algorithm,a weighted constrained Euclidean distance optimization strategy and triangular constrained optimization strategy based on SURF are proposed.The weighted constraint Euclidean distance optimization strategy can eliminate a large number of erroneous feature point pairs in the rough matching feature point pairs,improve the accuracy of the geometric transformation model and the efficiency of the image registration algorithm.The optimization strategy based on Delaunay triangulation divides the region of the image matching feature points into multiple triangle regions,and then matches the feature points in the triangle with corresponding correspondence.That is to say,this method can effectively increase the number of matching feature point pairs,thereby improving the accuracy of the image registration algorithm.In this paper,the images in the extreme illumination scenario are used as experimental data for image enhancement and image mosaic experiments,and the effectiveness and practicability of the optimization strategy proposed in this paper is verified by standard test images.