Research On Key Technologies Of Heterogeneous Image Registration

Image registration is a basic topic in the field of computer vision.Thanks to some excellent feature extraction,description and matching algorithms,at present,the ordinary problem of homogeneous image registration can be solved easily.Heterogeneous image registration is a type of problem in the image registration subject.It aims to research the registration problem of images acquired by multi-source imaging sensors.It is the necessary for multi-source image fusion and has important significance.However,many methods that can solve homogenous image registration problem cannot solve the heterogeneous image registration problem well due to the significant non-linear intensity differences caused by radiometric variations among such images,so the heterogeneous image registration problem is a very challenging subject.The content of this article is the heterogeneous image registration method,the main research object is the visible and infrared image registration.The main work of this article is as follows.1 This thesis introduces a scene monitoring project work that contains multi-view and multi-source information.The project work includes three modules: homogenous image stitching,heterogeneous image registration,and heterogeneous image fusion to meet the needs of multi-period and all-weather monitoring of large-field scenes.2.For some images containing large-scale open scenes,the image registration algorithm for key areas between heterogeneous images is constructed in this thesis.In this thesis,a key area extraction method based on saliency measurement and a template matching method based on the similarity of edge are implemented.The automatic registration algorithm of visible and infrared images including airport scenes reduces the redundant information in the images,and improves the accuracy of the algorithm,and reduces the computational complexity of the algorithm.3.In this thesis,a feature description method based on histogram of adaptive equalization second gradient is proposed and applied to visible and infrared image registration.This algorithm equalizes the infrared image gradient magnitude and computes the gradient of the equalized gradient magnitude image.The algorithm weakens the problem of the non-linear intensity differences between the visible and infrared images to a certain extent.In this thesis,this method is applied to the GLOH-like and SIFT feature description framework and combined with SIFT feature extraction algorithm to extract and describe feature points of visible and infrared images.Feature points matching results show that this method is superior to the traditional gradient histogram feature description algorithms and second-order gradient histogram feature description algorithms.