Adaptive Reliability-Calibrated Consensus-Complementarity-Conflict Modeling for Infrared and Visible Image Fusion

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared and visible image fusion aims to combine complementary information.
• Challenges include preserving visible texture and infrared saliency while avoiding information loss or distortion.
• Cross-modal differences can introduce noise and artifacts.

Purpose of the Study:

• To propose ARC3Fusion, a novel image fusion method.
• To achieve a better balance between visible texture details and infrared target saliency.
• To address information suppression, distortion, and artifacts in cross-modal fusion.

Main Methods:

• ARC3Fusion reformulates fusion as a reliability-calibrated consensus-complementarity-conflict process.
• It employs a progressive shared encoder and modality-specific residual adapters.
• Key steps include Cross-Modal Explainable Residual Decomposition, Trustworthy Complementarity Verification, Cross-Modal Conflict Estimation, and Conflict-Aware Routing.

Main Results:

• ARC3Fusion demonstrated consistent performance across LLVIP, MSRS, and TNO datasets.
• Achieved state-of-the-art performance on the LLVIP dataset with top scores in EN, SF, AG, VIF, and SCD.
• Verified modality-specific residuals and coordinated local conflicts improved joint preservation.

Conclusions:

• ARC3Fusion effectively balances visible texture details and infrared thermal saliency.
• The proposed verification and conflict coordination strategy enhances fusion quality.
• This approach offers a significant improvement over conventional fusion techniques.