Comparison of Different Image Fusion Methods Using Gaofen-6 and Sentinel-2 Imagery of Beijing City, China

Abstract

Pan-sharpening, a technique in image fusion, is gaining attention in the image processing community. It aims to merge low-spatial and high-spectral resolution images with high-spatial and low-spectral resolution images, resulting in high-spatial and high-spectral resolution images. However, multi-sensor image fusion often leads to spectral distortion in the fused images while the performance of fusion algorithms varies based on image characteristics. This study evaluates the performance of four pan-sharpening algorithms (Gram-Schmidt (GS), Color Normalized spectral (CNS), Principal Component (PC), and Nearest Neighbor Diffusion (NND)) by fusing images from two satellites: Sentinel-2 and Gaofen-6 of the Beijing city, China. The resulting pan-sharpened images are compared both qualitatively and quantitatively in terms of spectral fidelity, while spatial enhancements are assessed through visual interpretation. The comparative analysis of quantitative and qualitative approaches reveals that the GS algorithm produces highly comparable results for exhibiting high spectral quality and spatial enhancement of both sensors. The performance variation of CNS and PC methods for both sensors is relatively insignificant. However, the NND method demonstrated excellent spatial enhancement for the Gaofen-6 sensor while retrieving highly distorted radiometry images for the Sentinel-2 sensor.