DPAFuse: A dual-space probabilistic adversarial image fusion framework with robust coding embedding

Area of Science:

• Computer Vision
• Artificial Intelligence
• Signal Processing

Background:

• Existing multimodal image fusion methods struggle with composite degradations and modality bias.
• Current fusion strategies often lead to unstable representations due to enforced cross-modal consistency.
• This results in compromises and unreliable fused outputs.

Purpose of the Study:

• To propose DPAFuse, a dual-space probabilistic adversarial image fusion framework.
• To address the vulnerabilities of existing methods to composite degradations and modality bias.
• To achieve robust and high-fidelity multimodal image fusion.

Main Methods:

• Implemented a robust coding embedding mechanism to project heterogeneous inputs into a unified latent space.
• Developed a dual-space probabilistic adversarial fusion mechanism for adaptive global fusion.
• Enforced distribution-level consistency across latent and image domains to mitigate modality bias.

Main Results:

• DPAFuse demonstrated superior robustness against composite degradations.
• The framework achieved higher fusion fidelity compared to state-of-the-art methods.
• Experiments on public datasets validated the effectiveness of the proposed approach.

Conclusions:

• DPAFuse offers a robust solution for multimodal image fusion.
• The dual-space probabilistic adversarial approach effectively handles degradations and modality bias.
• The proposed method advances the field of image fusion with improved performance and stability.