A Pan-Sharpening Method based Latent Low-Rank Decomposition Model

Abstract

In this paper, we propose a novel method based on latent low-rank representation theory (LatLRD) for pansharpening, which aims to synthesize a high resolution multispectral (MS) image from a high resolution panchromatic (PAN) image and a low resolution MS image. Exploiting the property of the low-rank of the MS data, the LatLRD is first performed on the up-sampled MS image and the PAN image to reconstruct a composite image in order to preserve the spectral fidelity of MS images, while transferring spatial structures. Second, a multi-scale procedure is applied to the generated composite image from the LatLRD decomposition for extracting the spatial information. Finally, the details are injected to the up-sampled MS bands to obtain the corresponding MS image at fine resolution. Experimental results demonstrate that the proposed approach performs better than several state-of-the-art methods in enhancing the spatial quality and preserving the spectral fidelity.