A multi-directional fractional-order variation with luminance term for infrared and visible image fusion

Abstract

Infrared and visible image fusion strives to effectively combine the advantageous information from input images to achieve a fused result. In this paper, we propose a novel multi-directional fractional-order variation to fuse infrared-visible images. First, each fidelity term of each source image is modeled straightforwardly by Euclidean norm, ensuring robust optimization. Then, the detail regularization term is formulated based on fractional variation in two, four, and eight directions instead of integral variation, which enables the capture of more comprehensive detail while avoiding the undesirable staircase effect. Furthermore, the fused image is enhanced by transferring the highest level of brightness from the source images in the luminance regularization. Finally, based on the extensive experiments, the proposed method exhibits superior performance in both subjective and objective evaluations compared to existing others. Moreover, our method is further expanded to the multi-modal medical image fusion, achieving promising performance preliminarily.