A Novel Fidelity Based on the Adaptive Domain for Pansharpening

Jin-Liang Xiao, Ting-Zhu Huang, Liang-Jian Deng, Ting Xu

Abstract: Pansharpening aims to obtain the high resolution multispectral image (HRMS) using the panchromatic image (PAN) and low spatial resolution multispectral image (LRMS). The similarity between PAN and HRMS has shown powerful performance for spatial feature extraction. The prevailing methods usually describe the similarity on a fixed transformed domain. However, such domain, e.g., gradient domain, usually limits the preservation of spatial details and neglects flexibility. To overcome these challenges, we propose an adaptive transformed domain-based spatial fidelity to depict the similarity accurately and flexibly. Based on the proposed spatial fidelity, we build a novel variational pansharpening model that consists of spectral and spatial fidelity terms. We design an algorithm based on the alternating direction method of multiplier (ADMM) framework to solve the model. Experimental results on reduced- and full-resolution data verify the effectiveness of the proposed method.