Research On Multi-source Image Registration And Fusion Algorithm Based On Structural Features

With the rapid development of electronic technology and information technology,single sensor system has been eliminated,and multi-sensor system has been widely used.Among them,the infrared imaging equipment and the visible imaging equipment because of their inherent mechanism make infrared and visible images have natural complementarities.The combination of infrared image and visible image can not only describe the scene or the target clearly and accurately,but also can be applied to the subsequent processing.However,image registration is an essential preprocessing technique for image fusion,and the accuracy of registration directly affects the effect of fusion.This paper focuses on the research of the basic theory of image registration and image fusion,and the multi-source image registration algorithm based on graph is presented.In addition,on the basis of non-subsampled contourlet transform(NSCT),the fusion algorithm is improved,and on the basis of the existing fusion rules,the fusion rules are also improved.Firstly,the image registration technology is studied in this paper,considering that there may be a variety of changes between the original images,eventually the perspective transformation is selected as the image transform model.At the same time,it is difficult to extract common features,so the maximum stable extremal region(MSER)extraction algorithm is used to extract the features of infrared and visible images in this paper,the algorithm can extract the saliency regions in the image.In order to make full use of the extracted saliency regions,a new registration algorithm is presented in this paper based on the pyramid hierarchical and graph theory.First,extracting the MSERs in the determining maximum down-sampling images and describes the feature of each region.Then constructing the relation matrix of two graphs and optimizing the regional mapping relationship between infrared and visible images.Next according to the mapping relationship we can construct the initial point set for matching.Finally,using random sample consensus(RANSAC)algorithm to obtain the optimal parameters and determine the error evaluation parameters.In the high resolution image,current matching error as constraints for local exact match until the original resolution images,and the final transformation matrix can be obtained for final image registration.In addition,because the visible image contains a wealth of detailed information,so the secondary NSCT decomposition is proposed in this paper for the visible image.Based on regional fusion rules,the low and high frequency subbands have different fusion rules,in order to ensure that the fused image can retain all the useful information in the original image.The experimental results show that the proposed method can achieve high matching accuracy on the premise of robustness.At the same time,the fusion algorithm proposed in this paper can make the fusion result contain more abundant information.