Joint Spectral and Spatial Consistency Priors for Variational Pansharpening

Abstract

This paper proposes a new variational pansharpening model with joint spectral and spatial consistency priors, which aims to fuse a low resolution (LR) multispectral (MS) image and a high resolution (HR) panchromatic (Pan) image to produce a pan-sharpened HR MS image. Specifically, the proposed model combines three consistency terms into a unified variational framework, which are (1) Local spectral consistency fidelity term, which enforces the degradation relation-based local spectral consistency constraint between the HR MS and LR MS images; (2) Hessian feature-enforced spatial consistency prior term, which particularly models the Hessian feature consistency constraint between the HR MS and Pan images to enforce spatial consistency; and (3) Wavelet-based spectral-spatial consistency prior term, which models the consistency between the HR MS image and the constructed Wavelet-based matching image to enforce spectral-spatial consistency. Moreover, the proposed model is efficiently solved by designing an optimization algorithm under the forward-backward splitting framework. Finally, experiments on the QuickBird, Pleiades and GeoEye-1 satellite datasets systematically illustrate that the proposed method performs better spectral and spatial qualities than various compared methods.