Sparsity Regularization Based Spatial-Spectral Super-Resolution of Multispectral Imagery

摘要

Multispectral (MS) remote sensing image is composed of several spectral bands of distinct wavelengths. Most earth observation satellites provide MS images consisting several low-resolution (LR) bands together with a single high-resolution (HR) image. A single image super-resolution (SISR) method tries to produce a HR MS output from the given LR MS input using digital image processing algorithms. In this work, we present a patch-wise sparse representation based MS image SR using a coupled overcomplete trained dictionary. The dictionary learning is carried out from patches extracted from the given HR panchromatic (PAN) image itself. Experiments are carried out using test MS images from QuickBird satellites and results are compared with other state-of-the-art MS image SR and pan-sharpening methods.