DCBFusion: an infrared and visible image fusion method through detail enhancement, contrast reserve and brightness balance

Due to the complementary nature of visible and infrared images, they are widely used in image fusion to generate fused images containing more comprehensive information. Although existing fusion methods have achieved good results, there are some problems. In some cases, the features of an image are affected by a shot from another modality, which leads to the problem of background contamination and missing information. To solve these problems, we designed a visible and infrared image fusion network starting from three key factors that affect structural similarity. Our fusion network can avoid these problems through detail enhancement, contrast preservation, and luminance balancing. Through the cross-stage feature extraction and multi-scale feature enhancement modules achieve detail enhancement. The complementary information fusion module finds and fuses complementary information from different images to achieve contrast preservation. The loss function performs luminance balancing. Comparison and generalization experiments on several other public datasets show that our network effectively avoids background contamination and information loss and achieves outstanding results in both quantitative and qualitative aspects.