Recent advances via convolutional sparse representation model for pixel-level image fusion

Image fusion aims to integrate complementary information from different source images into the final output image. This plays a significant role in high-level vision tasks. However, image fusion methods based on sparse representation (SR) or conventional multiscale transform (MST) have some drawbacks that are difficult to overcome. As an alternative form of SR, convolutional sparse representation (CSR) has the advantages of detail preservation and shift-invariance, which can overcome the shortcomings of SR- and MST-based fusion methods. Since CSR has been widely used in the field of image fusion and has advanced this field to a great extent, it is necessary to conduct a comprehensive investigation of image fusion based on CSR. To the best of our knowledge, there are no previous papers reviewing and evaluating CSR-based fusion methods, and this study is the first retrospective. In this paper, we focus on CSR-based image fusion methods and review the recent advances in pixel-level image fusion based on CSR. In the experimental part of the paper, multifocal images, infrared-visible images, and multimodal medical images are used as test images to compare and evaluate the performance of different image fusion methods. In addition, the future trend of CSR-based image fusion is discussed. This paper is expected to serve as a resource of reference for both researchers and general learners seeking an overview of CSR-based image fusion.