Diff-IF: Multi-modality image fusion via diffusion model with fusion knowledge prior

Xunpeng Yi, Linfeng Tang, Hao Zhang, Han Xu, Jiayi Ma

Abstract: Highlights•We propose a novel trainable diffusion model for multi-modal image fusion.•It provides a feasible paradigm for diffusion model in lacking ground truth.•It avoids unstable training of GAN-based fusion methods.•It has better flexibility and less cost than existing diffusion-based methods.•Our method exhibits good fusion results and excellent semantic representation.