Boosting the Quality of Pansharpened Image by Adjusted Anchored Neighborhood Regression

Abstract

Pansharpening technology integrates low spatial resolution (LR) multi-spectral (MS) image and high spatial resolution panchromatic (PAN) image into a high spatial resolution multi-spectral (HRMS) image. Various pansharpening methods have been proposed, and each of them has its own improvements in different aspects. Meanwhile, there also exist specified shortages within each pansharpening method. For example, the methods based on component substitution (CS) always cause color distortion and multi-resolution analysis (MRA) based methods may loss some details in PAN image. In this paper, we proposed a quality boosting strategy for the pansharpened image obtained from a given method. The A+ regressors learned from the pansharpened results of a certain method and the ground-truth HRMS images are used to overcome the shortages of the given method. Firstly, the pansharpened images are produced by ATWT-based pansharpening method. Then, the projection from the pansharpened image to ideal ground truth image is learned with adjusted anchored neighborhood regression (A+) and the learned A+ regressors are used to boost quality of pansharpened image. The experimental results demonstrate that the proposed algorithm provides superior performances in terms of both objective evaluation and subjective visual quality.