Image Fusion by Hierarchical Joint Sparse Representation

Joint sparse representation (JSR) based image fusion, as one of competitive sparse representation based fusion methods, has been widely studied recently. In this kind of methods, image features are represented as sparse coefficients. They are typically calculated with two decomposition algorithms, namely orthogonal matching pursuit and basis pursuit. In both of them, an error tolerance parameter is specified to control the fineness of a fused image. Intuitively, the more detailed an image fineness is, the more micro-information is presented; the more rough it is, the more macro-information is summarized. Therefore, it is reasonable to assume that complementary information exists among the images generated by different error tolerance parameters. Motivated by this, in this paper, we have tried to combine the features in these images and verify the above assumption. Specifically, we have proposed a two-layer hierarchical framework based on JSR. Extensive experiments demonstrate that effectively combining features in images of different fineness does improve the quality of the fused image significantly. The proposed framework outperforms previous methods according to many objective evaluation criteria.