Remote sensing image fusion based on morphological filter and convolutional sparse representation

Abstract

Remote sensing image fusion is the process of obtaining high-resolution multispectral images by fusing spectral information-rich multispectral images and spatial information-rich panchromatic images. Sparse representation has achieved good results in this field, but the sparse representation is encoded in blocks, which destroys the correlation between image blocks and thus causes the problems of spectral distortion and missing details in the fusion results. To address the above problems, a fusion algorithm combining convolutional sparse representation and morphological filter is proposed. The convolutional sparse representation can represent the whole image sparsely, which fully considers the correlation between pixels and reduces the spectral distortion of the fusion results. The morphological filter can estimate the spatial details of the image more accurately, so that more spatial detail information can be obtained in the fusion result. And the methods based on multiplicative injection is used, aiming to inject more detailed information into the fusion results. The experimental results show that the objective evaluation index of the fusion results obtained by this method is better, and the subjective visual effect is better.