Infrared And Visible Image Fusion Method Based On Sparse Representation

As a hot branch of the field of multi-source image fusion,infrared and visible image fusion technology plays an important role in military,biology,monitoring industry and other fields.The research on the fusion of infrared and visible images with complementary feature has a comparative significance.In this thesis,some image fusion algorithms based on sparse representation have been achieved.Then the self-adaptation image fusion algorithm model is proposed based on local energy and variance to implement infrared and visible image fusion on the basis of traditional sparse fusion algorithm framework.The main work and results are shown as follows:(1)The basic principle of sparse representation has been studied.Then we discuss the factors that affect the performance of the fusion algorithms,such as sliding window size,sliding step size and fusion rules,and simulation analysis is carried out.Through the comparison and analysis of the fusion results,the optimum values of each parameter of the same type of sparse fusion algorithm is determined.(2)Based on the research of the fusion rule of the sparse fusion algorithm and the characteristics of the infrared and visible image,a new image fusion method is proposed based on local energy and variance.Each image block is classified to select the fusion rule adaptively to get the final fusion image.The simulation and performance analysis of several fusion algorithms show that the proposed method is feasible and effective.(3)The conventional quality evaluation methods of image fusion have been studied.According to the scene of images and the purpose of fusion,the fusion evaluation indexes are classified on the basis of the existing objective evaluation methods,which provides guidance for the selection of performance evaluation of the fused image.(4)Multiple sets of infrared and visible video sequences is fused by the proposed algorithm for target detection.Compared with image sequences acquired by single sensor for target detection,the fused image sequences obtained by the proposed method has a better detection effect.