A full-scale hierarchical encoder-decoder network with cascading edge-prior for infrared and visible image fusion

Xiaoqing Luo, Juan Wang, Zhancheng Zhang, Xiao-Jun Wu

Abstract: Highlights•A novel full-scale hierarchical encoder-decoder network is proposed to unify the full-scale long-range semantic information and single-scale shallow details of multi-modal images.•To comprehensively utilize the multilayer effective information and inject the edge prior, a triple fusion mechanism is proposed based on dual-attention fusion (DAF) strategy.•A cascading edge-prior branch is designed to jointly guide the decoder focusing on abundant details layer-by-layer.•A novel loss function consisting of SSIM loss, intensity loss and edge loss is constructed to further maintain the network with better edge representation and reconstruction capability.