Dictionary learning based pan-sharpening

Dehong Liu, Petros T. Boufounos

Abstract: Pan-sharpening is an image fusion process in which high resolution (HR) panchromatic (Pan) imagery is used to sharpen the corresponding low resolution (LR) multi-spectral (MS) imagery. Pan-sharpened MS images generally have high spatial resolutions, but exhibit color distortions. In this paper, we propose a dictionary learning based pan-sharpening process to reduce the color distortion caused by the interpolation of the MS imagery. Instead of interpolating the LR MS image before fusion, we generate an improved MS image which is sparse with respect to a dictionary learned from the image data. Our experiments on degraded QuickBird and IKONOS images demonstrate that the distortion in the MS images produced using our approach is significantly reduced.