A Two-Branch Infrared and Visible Image Fusion Network Based on Attention Guidance

Infrared and visible image fusion aims to enhance image quality by integrating the complementary information of the two images. Due to the significant differences between the two modalities, it is easy to ignore the effective interaction between global and detail information during the information fusion process, resulting in the loss of some important contextual information. Therefore, in this paper, we propose an attentionguided two-branch fusion network (AGFuse) based on attention guidance. Specifically, we design a residual attention module (RAM) for shallow information extraction. Then, we introduce a two-branch feature extractor, where the base encoder extracts higher-level global features by combining convolution with a selfattention mechanism, while the detail encoder captures fine information in the image by separating and fusing detailed features. Further, we introduce the CGAFusion module, which weightedly fuses the global and detail features extracted by the encoder and outputs the fused image after optimized processing by the Transformer Block. A large number of experiments on mainstream data sets show that the proposed method is superior to other representative advanced methods.