DFGIC-Net: diffusion feature-guided information complementary network for infrared and visible light fusion

Yekai Cui, Peng Duan, Jinjiang Li

Abstract: This paper addresses the challenge of preserving detailed information, such as target edges and textures, in multimodal image fusion, particularly for infrared and visible images. We propose the Diffusion Feature Guidance Network based on Information Complementarity (DFGIC-Net), which leverages diffusion features for enhanced multimodal fusion. The network consists of two main components: first, a strategy to generate diffusion features that balance multimodal information and enhance edge details, and second, a wavelet transform-based approach to decompose image features into high- and low-frequency components, which are then optimized through an Information Complement Enhancement Module (ICEM). This module bridges the semantic gap between modalities, ensuring comprehensive fusion of frequency and spatial domain information. Our Fusion-Guide Head combines these features with a learnable boundary enhancement strategy to produce fused images with sharp boundaries and high fidelity. Extensive experiments show that DFGIC-Net outperforms state-of-the-art fusion methods in multiple metrics and retains superior performance in downstream tasks, providing strong support for advanced visual applications. Our code will be available at https://github.com/ccnokk/DFGIC.

External IDs:dblp:journals/tjs/CuiDL25