FDIFusion: A fusion-decoupling framework with implicit neural representation for infrared and visible image fusion

Abstract

Infrared and Visible Image Fusion (IVIF) aims to combine the complementary advantages of infrared and visible images to produce high-quality fused images with enhanced clarity and information content. Traditional methods often fail to independently process infrared and visible images, thereby missing the opportunity to fully exploit their complementary information. To address this issue, we propose an innovative framework that employs implicit neural representation (INR) for progressive fusion and decoupling. Our framework incorporates a triple-branch, multi-resolution network designed to minimize information loss during fusion and effectively extract features from infrared and visible modalities. By integrating a decoupling model, our method reduces redundancy and enhances the independence of fused features. Furthermore, INR enables continuous feature representation across network layers, which enhances structural continuity and semantic consistency in the fused image. This, in turn, facilitates the preservation of fine details and improves visual perception. Experimental results demonstrate that our proposed FDIFusion method strikes an optimal balance between salient structures and fine textures, achieving significant quantitative and qualitative improvements over existing methods. In addition, FDIFusion exhibits excellent generalization ability and has great potential for real-world applications in complex scenarios.