Compressed Sensing based pansharpening technique with learned dictionary

In this paper we propose a new pan-sharpening method using compressed sensing (CS) with learned dictionary. Given the low-resolution multispectral (MS) and a high-resolution panchromatic (PAN) images, pan-sharpening is a algorithmic technique to combine them into single image with high resolution multispectral (HMS) image. We model the given low spatial resolution MS image as a measurement projection in the CS theory. First, we constructs a dictionary which is learned using MS and PAN images. Next, the learned dictionary is used to estimate the sparse vector in the CS theory where CoSaMP is used. We tested the potential of the proposed method by conducing the experiments on the datasets of the Quickbird and Worldview-2 satellites. In addition, the quantitative analysis in terms of different traditional measures are also evaluated. The results and the measures indicate that the proposed method exhibit better fusion when compared to other pan-sharpening techniques.