Underwater Image Enhancement with Physical-based Denoising Diffusion Implicit Models

Nguyen Gia Bach,Chanh Minh Tran,Eiji Kamioka,Phan Xuan Tan

2024-09-27

Abstract:Underwater vision is crucial for autonomous underwater vehicles (AUVs), and enhancing degraded underwater images in real-time on a resource-constrained AUV is a key challenge due to factors like light absorption and scattering, or the sufficient model computational complexity to resolve such factors. Traditional image enhancement techniques lack adaptability to varying underwater conditions, while learning-based methods, particularly those using convolutional neural networks (CNNs) and generative adversarial networks (GANs), offer more robust solutions but face limitations such as inadequate enhancement, unstable training, or mode collapse. Denoising diffusion probabilistic models (DDPMs) have emerged as a state-of-the-art approach in image-to-image tasks but require intensive computational complexity to achieve the desired underwater image enhancement (UIE) using the recent UW-DDPM solution. To address these challenges, this paper introduces UW-DiffPhys, a novel physical-based and diffusion-based UIE approach. UW-DiffPhys combines light-computation physical-based UIE network components with a denoising U-Net to replace the computationally intensive distribution transformation U-Net in the existing UW-DDPM framework, reducing complexity while maintaining performance. Additionally, the Denoising Diffusion Implicit Model (DDIM) is employed to accelerate the inference process through non-Markovian sampling. Experimental results demonstrate that UW-DiffPhys achieved a substantial reduction in computational complexity and inference time compared to UW-DDPM, with competitive performance in key metrics such as PSNR, SSIM, UCIQE, and an improvement in the overall underwater image quality UIQM metric. The implementation code can be found at the following repository: <a class="link-external link-https" href="https://github.com/bachzz/UW-DiffPhys" rel="external noopener nofollow">this https URL</a>

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to address the challenges of real - time processing of underwater image enhancement (UIE) on autonomous underwater vehicles (AUVs). Specifically, the paper focuses on the following key issues:

1. **Underwater image degradation**:
Underwater images are severely degraded due to factors such as light absorption and scattering, resulting in problems such as poor image quality, color distortion, and low contrast. This seriously affects the AUV's ability to perform tasks.

2. **Limitations of existing methods**:
- **Traditional methods**: They rely on prior knowledge or assumptions, have poor adaptability, and are prone to over - enhancement or under - enhancement results.
- **Convolutional neural network (CNN) - based methods**: Although they have good performance, their generalization ability is limited under different underwater conditions.
- **Generative adversarial network (GAN)**: It has unstable training, mode collapse, and insufficient diversity of generated samples.
- **Denoising diffusion probability model (DDPM)**: Although it performs well in image - to - image tasks, it has high computational complexity, long inference time, and is difficult to achieve real - time processing.

3. **Limited computing resources**:
AUVs usually have limited computing resources, and the existing DDPM methods (such as UW - DDPM) have too high computational complexity to achieve real - time image enhancement on these devices.

To solve the above problems, the paper proposes a new underwater image enhancement method that combines physics and diffusion - **UW - DiffPhys**. The main objectives of this method are:

- **Reduce computational complexity**: By introducing lightweight physical model components and a denoising U - Net, replace the original distribution - transformation U - Net with high computational complexity.
- **Accelerate the inference process**: Adopt the non - Markovian sampling technique (DDIM) to reduce the number of inference steps and speed up the processing speed.
- **Maintain or improve the enhancement effect**: Remain competitive in key indicators such as PSNR, SSIM, UCIQE, and significantly improve the overall underwater image quality (UIQM).

Through these improvements, UW - DiffPhys can achieve more efficient underwater image enhancement on low - power AUVs and meet the requirements of real - time processing.