BlurDM: A Blur Diffusion Model for Image Deblurring

2025-12-03

AI Summary

Existing diffusion-based methods for dynamic scene image deblurring neglect the underlying physical mechanism of motion blur formation, leading to suboptimal performance. To address this, we propose BlurDM—the first approach to implicitly model the continuous exposure nature of motion blur as a dual diffusion forward process in latent space, jointly diffusing noise and blur. This enables coupled denoising and deblurring during reverse inference. BlurDM incorporates the dual-diffusion mechanism as a flexible, physics-informed prior into the deblurring network, conditioned on Gaussian noise. Evaluated on four mainstream benchmarks, BlurDM consistently outperforms state-of-the-art methods, achieving average improvements of 1.2–2.4 dB in PSNR and 0.015–0.028 in SSIM. These results demonstrate its effectiveness, generalizability, and physical interpretability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Mixed Discrete/Continuous Search
• Machine Learning: Life-Long and Continual Learning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Diffusion models show promise for dynamic scene deblurring; however, existing studies often fail to leverage the intrinsic nature of the blurring process within diffusion models, limiting their full potential. To address it, we present a Blur Diffusion Model (BlurDM), which seamlessly integrates the blur formation process into diffusion for image deblurring. Observing that motion blur stems from continuous exposure, BlurDM implicitly models the blur formation process through a dual-diffusion forward scheme, diffusing both noise and blur onto a sharp image. During the reverse generation process, we derive a dual denoising and deblurring formulation, enabling BlurDM to recover the sharp image by simultaneously denoising and deblurring, given pure Gaussian noise conditioned on the blurred image as input. Additionally, to efficiently integrate BlurDM into deblurring networks, we perform BlurDM in the latent space, forming a flexible prior generation network for deblurring. Extensive experiments demonstrate that BlurDM significantly and consistently enhances existing deblurring methods on four benchmark datasets. The source code is available at https://github.com/Jin-Ting-He/BlurDM.

Problem

• Integrates blur formation into diffusion for deblurring
• Models dual-diffusion to denoise and deblur simultaneously
• Enhances existing methods via latent space prior generation

Innovation

• Integrates blur formation into diffusion model
• Uses dual-diffusion for noise and blur modeling
• Performs latent space processing for efficiency