Research On Multi-source Image Fusion Based On Low-rank Decomposition And Convolutional Sparse Coding

In many fields of modern society,multiple sensors are often used to capture multiple source images of the same scene.These source images contain different information.The task of image fusion is to fuse different information obtained by multiple sensors,so that the information between source images can be complementary and integrated into one image.The fused image will be more beneficial to the majority of scholars Further research and application in other industries.For the image fusion task,the previous sparse representation method blocks the source image,but this method will damage some structure and feature information of the image to a certain extent,and affect the whole structure characteristics of the image.The image fusion model based on convolutional sparse coding is not restricted by the dimension of image,so it can make up for the defect of image fusion model based on block.Based on the advantages of convolutional sparse coding that can better maintain image feature information,the thesis studies an image fusion method based on low-rank decomposition and convolutional sparse coding,and designs a filter model.The main innovative work is as follows:(1)Aiming at the problem that the traditional sparse representation fusion method blocks the image and affects the image structure,the thesis makes a global processing for each source image.Firstly,the low rank decomposition of the source image is carried out to obtain the low rank and sparse parts,and then the low rank components and sparse components are fused by different fusion rules.Finally,the fused image is obtained by adding the low rank and sparse components.(2)Due to the different information contained in different image components,the thesis designs different fusion rules for the fusion of the two image components.For the fusion of sparse components,the thesis designs a filter model,trains the filter dictionary,decomposes the sparse components through the filter dictionary,and obtains the corresponding sparse feature map,and then adopts an improved fusion based on Laplace energy sum The rule gets the fusion sparse feature map,and finally the fusion sparse component is reconstructed.For the fusion of low-rank components,the thesis adopts fusion rules based on image brightness information to obtain fused low-rank components.Through fusion experiments on three types of images: medical image,multi-focus image,infrared and visible light image,the experimental results verify the effectiveness of the method in the thesis.