Skip to content
Book Open access

Infrared-Visible Image Fusion with Dual Degradation-Aware Priors

Nov 2026 · Proceedings of the 34th ACM International Conference on Multimedia · 0 citations · 42 references

Abstract

Degraded infrared-visible image fusion (DIVF) aims to integrate multimodal images under degradation factors. Existing methods typically address a single degradation type, limiting generalizability. Recent approaches have adopted large language models (LLMs) to simulate user instructions, yielding linguistic priors to handle diverse degradations universally. Despite promising results, they still face three issues: (1) semantic homogenization, as monotonic generation rules for LLMs restrict the diversity of generated prompts; (2) semantic ambiguity, as vague user instructions make it hard for the model to identify degradation types; and (3) detail loss, as high-level linguistic priors often overlook low-level cues crucial for preserving fine-grained visual details. In this paper, we propose DDAP, a novel text-guided universal DIVF framework that integrates Dual Degradation-Aware Priors to address diverse degradation types and user instructions. We first outline three guiding principles for dual priors, ensuring their coordination and adaptive perception of different degradations. For the linguistic prior, we introduce three-tier generation rules to enhance semantic diversity in LLM-generated prompts, along with a Linguistic-prior Adaptive Calibration (LAC) module to refine textual embeddings for enhanced interaction. For the frequency prior, we design the Frequency-prior Adaptive Decoupling (FAD) module to capture implicit degradation patterns based on subband distribution, enhancing detail preservation. Finally, we propose a Dual-Prior Fusion module to integrate both priors and apply an auxiliary classification task to further ensure degradation-awareness. Extensive experiments on degraded, compound, and unseen real-world scenarios validate the superiority and generalizability of our DDAP framework.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.