ETBIFusion: An infrared and visible image fusion network with edge-texture enhancement and bidirectional interaction

Deep learning-based methods for infrared and visible image fusion encounter significant challenges in effectively managing detailed information and feature representation, particularly due to the risk of detail loss and insufficient feature expression during convolution operations. To address these issues, an infrared and visible image fusion network with edge-texture enhancement and bidirectional interaction is proposed. The framework includes a progressive multi-branch (PMB) module, a bidirectional feature interaction fusion (BFIF) module, and a difference-aware edge-texture (DAET) module. Specifically, the PMB effectively extracts multi-level feature information through layered convolution, which solves the issue of detail loss in convolution operations and significantly improves the detailed feature extraction of infrared and visible images. The BFIF optimizes the alignment and fusion process of cross-modal features through a bidirectional interaction mechanism and enhances the spatial and contextual information of the fused images. The DAET combines edge detection and texture enhancement technology, which further improves the detail expression and texture details of the fused images and ensures that the details and overall quality of the generated images have been significantly improved. Finally, the optimization of the loss function further enhances the edge presentation and reconstruction capabilities. Extensive experiments on the MSRS, TNO, and RoadScene datasets demonstrate that ETBIFusion achieves the best performance in at least three of the six evaluation metrics compared to nine other state-of-the-art (SOTA) image fusion methods.