A Variational Pansharpening Algorithm Based on Total Variation and Primal-Dual Optimization

Abstract

This paper proposes a variational framework to estimate the high-resolution (HR) multispectral (MS) image from the low-resolution (LR) MS image and the panchromatic (Pan) image. The LR MS image is modeled as a decimation of the HR MS image. Furthermore, the Pan image is considered as a linear combination of the HR MS bands. A super-resolution (SR) model is defined in accordance with the image observation model and the total variation (TV) regularization. The SR reconstruction problem is modeled as a minimization problem, which is solved by an efficient primal-dual algorithm in a Euclidean setting. The result of comparing the proposed method with some recent classical and variational pansharpening methods proves the superiority of the proposed variational pansharpening algorithm.