Attention-Driven Networks for Edge-Preserving Infrared and Visible Image Fusion

An Attention-Driven Network for Edge-Preserving Infrared and Visible Image Fusion is proposed in this study, based on deep learning. In the proposed method, a dual-branch Transformer architecture is employed to separately extract base and detail features from the infrared and visible modalities. To enhance feature reuse capability, Gradient Residual Dense Blocks (GRDBs) are incorporated into the network. This design is intended to address the limitations of existing approaches in cross-modal feature fusion and detail preservation. Experimental results demonstrate that the proposed method outperforms state-of-the-art algorithms on the LLVIP dataset in terms of both subjective visual quality and objective evaluation metrics (EN, SD, VIF, etc.), significantly improving the quality of fused images.