Refined Pan-Sharpening With NSCT and Hierarchical Sparse Autoencoder

Most of available pan-sharpening technologies suffer from spectral and spatial distortions, for the coarse extraction from Panchromatic (Pan) image and brute injection of details to multispectral (MS) images. In this paper, in order to reduce the color distortion and enhance the spatial information of fused images, we propose a refined pan-sharpening (RPS) method using geometric multiscale analysis (GMA) and hierarchical sparse autoencoder (HSAE). First, a GMA tool, nonsubsampled contourlet transform (NSCT), is used to capture directional details of the Pan image at multiple scales. Then at each scale, HSAE is developed to gradually filter out the refined spatial details, via sparsely coding details under spatial self-dictionaries. The refined details are then injected into MS images to alleviate spectral distortions. By exploring the spatial structure in images and refining the spatial details injection via HSAE, RPS can reduce distortions to present fidelity colors and sharp appearance. Some experiments are taken on several datasets collected by QuickBird, Geoeye, and IKONOS satellites, and the experimental results show that RPS can reduce distortions in both the spectral and spatial domains, and outperform some related methods in terms of both visual results and numerical guidelines.