DDFusion:Degradation-Decoupled Fusion Framework for Robust Infrared and Visible Images Fusion

TL;DR

DDFusion tackles the realistic challenge of infrared–visible image fusion under compound degradations by decoupling degradation effects from fusion. It introduces DDON to perform modality-specific degradation separation (2D-DCT for infrared, Retinex for visible) and extract degradation-aware features, and ILGFN to fuse these features through multi-scale local and global interactions. The framework is trained with a degradation-focused loss and a fusion loss that jointly preserve structure, texture, and perceptual content. Extensive experiments show superior performance over state-of-the-art methods on degraded and non-degraded data, with demonstrated gains in downstream tasks such as object detection, highlighting practical relevance for robust IVIF systems.

Abstract

Conventional infrared and visible image fusion(IVIF) methods often assume high-quality inputs, neglecting real-world degradations such as low-light and noise, which limits their practical applicability. To address this, we propose a Degradation-Decoupled Fusion(DDFusion) framework, which achieves degradation decoupling and jointly models degradation suppression and image fusion in a unified manner. Specifically, the Degradation-Decoupled Optimization Network(DDON) performs degradation-specific decomposition to decouple inter-degradation and degradation-information components, followed by component-specific extraction paths for effective suppression of degradation and enhancement of informative features. The Interactive Local-Global Fusion Network (ILGFN) aggregates complementary features across multi-scale pathways and alleviates performance degradation caused by the decoupling between degradation optimization and image fusion. Extensive experiments demonstrate that DDFusion achieves superior fusion performance under both clean and degraded conditions. Our code is available at https://github.com/Lmmh058/DDFusion.

Figures (11)

• Figure 1: Compound degradation image fusion. pre. denotes degradation-specific pre-processing.
• Figure 2: Degraded infrared and visible images are decomposed via frequency and Retinex methods: visible images yield separated detail and degraded luminance, while infrared images reveal Gaussian and stripe noise.
• Figure 3: Overall architecture of the proposed DDFusion. The DDFusion framework comprises two principal subnetworks: the Degradation-Decoupled Optimization Network (DDON) and the Interactive Local-Global Fusion Network (ILGFN).
• Figure 4: Architecture of Interactive Transformer-based Block(ITB)
• Figure 5: Architecture of Residual Depthwise Separable Convolution Block(RDSCB), where $n$ denotes the depthwise convolution kernel size and $m$ indicates the repetition count of the module within the gray block.