Research On Image Fusion Algorithm Based On Sparse Representation

Image fusion technology can make up for the lack of a single sensor, and obtain more reliable, accurate and comprehensive image data. Image fusion technology has been applied to many aspects, such as, remote sensing, military, medical and medical treatment and so on. On the one hand, at present, the research of image fusion was mainly concentrated on the fusion method based on image sparse representation, including based on the multiscale decomposition and fusion method based on redundant dictionary decomposition, but the problems exist in two methods. in order to solve these problems, this paper try to find a better image fusion method. On the oher hand, at present, the choice of image fusion objective quality assessment is usually based on the experience, lacking of study on how to choose a reasonable objective evaluation index set. Therefore, this paper will try to find a reasonable choice of image fusion objective evaluation index selection.In this paper, Firstly, we study the fusion method based on multiscale decomposition, consist of wavelet transform(DWT), curvelet transform(CVT), nonsubsampled contourlet transform(NSCT) and nonsubsampled shearlet transform(NSST), by comparing and analysis the experimental result, we find the selection criteria of the number of decomposition layers and filters. At the same time, NSST has been got the best performance.Secondly, we study the image fusion method based on the branch dictionary, and prove that SR can improve the robustness of the fusion system. Finally, based on the two algorithms, through experimental analysis this paper find the advantages and disadvantages of the two algorithms, and found they have complementary characteristics, so we propose a new image fusion algorithm based on NSST and SR. The first of the image were NSST decomposition, and in low frequency with the learned dictionary, the low-frequency image sparse representation, choose 1 maximum norm method fusion, to regional energy is a significant indicator of fusion belt in the high frequency sub bands. Finally, through the NSST inverse transform, the fusion result is obtained. For the proposed algorithm, this paper then through the multi-focus image, infrared visible image, and medical image verify the advantages of the algorithm. The research of this paper proves that the image fusion algorithm based on NSST and SR can fully retain its advantages and make up its deficiency.Research on objective evaluation index set for image fusion. In this paper, multi-focus image fusion is used as the application background, and an objective evaluation index set of image fusion is proposed. In this method, we first analysis the correlation of image fusion objective evaluation index. Then, classify these indexs based on the correlation. Then, according to the specific application of the multi-focus image, the consistency of the index has been analyzed, through the clear image, noise image and non-registeriion image. The experimental results show that the fusion set of the objective evaluation index of the multi focus image can be obtained by the above method.