ID-Blau: Image Deblurring by Implicit Diffusion-based reBLurring AUgmentation

Jia-Hao Wu,Fu-Jen Tsai,Yan-Tsung Peng,Chung-Chi Tsai,Chia-Wen Lin,Yen-Yu Lin

2024-05-21

Abstract:Image deblurring aims to remove undesired blurs from an image captured in a dynamic scene. Much research has been dedicated to improving deblurring performance through model architectural designs. However, there is little work on data augmentation for image deblurring. Since continuous motion causes blurred artifacts during image exposure, we aspire to develop a groundbreaking blur augmentation method to generate diverse blurred images by simulating motion trajectories in a continuous space. This paper proposes Implicit Diffusion-based reBLurring AUgmentation (ID-Blau), utilizing a sharp image paired with a controllable blur condition map to produce a corresponding blurred image. We parameterize the blur patterns of a blurred image with their orientations and magnitudes as a pixel-wise blur condition map to simulate motion trajectories and implicitly represent them in a continuous space. By sampling diverse blur conditions, ID-Blau can generate various blurred images unseen in the training set. Experimental results demonstrate that ID-Blau can produce realistic blurred images for training and thus significantly improve performance for state-of-the-art deblurring models. The source code is available at

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the issue of data augmentation in image deblurring. Specifically, existing image deblurring methods mainly focus on the design of model architectures, while research on data augmentation is relatively scarce. Since continuous motion during image exposure can cause blur artifacts, the authors aim to develop an innovative blur augmentation method that generates diverse blurred images by simulating motion trajectories in continuous space.

### Main Contributions of the Paper

1. **Proposing ID-Blau**: A stable and controllable blur augmentation strategy for enhancing image deblurring in dynamic scenes.
2. **Modeling Continuous Blur Condition Fields**: Implicitly representing blur direction and intensity, allowing the sampling of various pixel-level blur condition maps to generate diverse reblurred images not seen in the training set.
3. **Integrating Pixel-Level Blur Condition Maps**: Combining pixel-level blur condition maps with a diffusion model to generate high-quality reblurred images.
4. **Experimental Results**: Experimental results show that ID-Blau significantly improves the performance of existing deblurring models and outperforms state-of-the-art methods in data augmentation.

### Method Overview

- **Blur Conditions**: Representing blur through direction and intensity in continuous space to generate blur condition maps.
- **ID-Blau Model**: Based on a conditional diffusion model, it inputs clear images and blur condition maps to generate blurred images.
- **Training and Inference Process**: Optimizing the model through forward diffusion and backward denoising processes to generate diverse blurred images.

### Experimental Results

- **Quantitative Results**: Experiments on the GoPro, HIDE, and RealBlur datasets show that ID-Blau significantly enhances the performance of multiple deblurring models.
- **Qualitative Results**: Visual quality comparisons show that ID-Blau significantly improves deblurring effects.
- **Ablation Studies**: Analyzing the impact of different blur directions and intensities on deblurring performance, validating the effectiveness and robustness of ID-Blau.

### Conclusion

This paper proposes an innovative blur augmentation method, ID-Blau, which significantly enhances the performance of image deblurring models by generating diverse blurred images. This method not only demonstrates technical innovation but also shows strong potential in practical applications.