A pixel-level image fusion framework based on structural similarity

This paper introduces a novel pixel-level image fusion framework based on structural similarity (SSIM). SSIM is an image quality assessment metric developed recently through comparing local patterns of pixel intensities from luminance, contrast and structure. In our scheme, the relationship of input images is classified three kinds of cases by contrasting the SSIM value of the original images with two thresholds, a lower limit and an upper limit. Then the fusion rule can be respectively determined as maximum selection, image assimilation, and block selection according to the different relationship. In addition, a image assimilation method is revealed by ascending gradient of SSIM. The proposed scheme is implemented within some different sets of images. Simulation results show that our framework provides promising fusion performance with good perceive.