A Regularized Model-Based Optimization Framework for Pan-Sharpening

Pan-sharpening is a common postprocessing operation for captured multispectral satellite imagery, where the spatial resolution of images gathered in various spectral bands is enhanced by fusing them with a panchromatic image captured at a higher resolution. In this paper, pan-sharpening is formulated as the problem of jointly estimating the high-resolution (HR) multispectral images to minimize an objective function comprised of the sum of squared residual errors in physically motivated observation models of the low-resolution (LR) multispectral and the HR panchromatic images and a correlation-dependent regularization term. The objective function differs from and improves upon previously reported model-based optimization approaches to pan-sharpening in two major aspects: 1) a new regularization term is introduced and 2) a highpass filter, complementary to the lowpass filter for the LR spectral observations, is introduced for the residual error corresponding to the panchromatic observation model. To obtain pan-sharpened images, an iterative algorithm is developed to solve the proposed joint minimization. The proposed algorithm is compared with previously proposed methods both visually and using established quantitative measures of SNR, spectral angle mapper, relative dimensionless global error in synthesis, Q, and Q4 indices. Both the quantitative results and visual evaluation demonstrate that the proposed joint formulation provides superior results compared with pre-existing methods. A software implementation is provided.