Compressive Sensing for Pan Sharpening

Abstract

In the mid 1980s, pan-sharpening caught significant attention in remote sensing and image processing, as SPOT 1 (launched in 1986) provided high resolution (HR) pan images and corresponding low resolution (LR) multispectral images. Since that time, intensive research has been carried out to develop efficient methods for generating fused images with both high spatial and spectral resolution. Compressive sensing (CS) is a state-of-the-art signal processing technique. Its super-resolution capability and robustness have been demonstrated. This article presents a new sophisticated pan-sharpening method based on CS. Visual and statistic analysis show superior performance of the proposed method compared to the existing conventional pan-sharpening methods in general, i.e. rich in spatial information and less spectral distortion. Moreover, popular quality assessment metrics are employed to explore the dependency on regularization parameters and to compare the efficiency of various sparse reconstruction toolboxes. In addition, for a more general application, i.e. no HR pan image is available, the proposed algorithm is extended by introducing a beforehand co-trained global over-complete dictionary pair.