Assessment of the SFIM algorithm

Abstract

Fusion of images with different spatial and spectral resolutions can improve the visualization of the images. Many fusion techniques have been developed to improve the spectral fidelity and/or spatial texture quality of fused imagery. Of them, a recently proposed algorithm, the SFIM (Smoothing Filter-based Intensity Modulation), is known for its high spectral fidelity and simplicity. However, the study and evaluation of the algorithm were only based on spectral and spatial criteria. Therefore, this paper aims to further study the classification accuracy of the SFIM-fused imagery. Three other simple fusion algorithms, High-Pass Filter (HPF), Multiplication (MLT), and Modified Brovey (MB), have been employed for further evaluation of the SFIM. The study is based on a Landsat-7 ETM+ sub-scene covering the urban fringe of southeastern Fuzhou City of China. The effectiveness of the algorithm has been evaluated on the basis of spectral fidelity, high spatial frequency information absorption, and classification accuracy. The study reveals that the difference in smoothing filter kernel sizes used in producing the SFIM-fused images can affect the classification accuracy. Compared with three other algorithms, the SFIM transform is the best method in retaining spectral information of the original image and in getting best classification results.