A unified framework of non-local parametric methods for image denoising

Sébastien Herbreteau,Charles Kervrann

2024-02-21

Abstract:We propose a unified view of non-local methods for single-image denoising, for which BM3D is the most popular representative, that operate by gathering noisy patches together according to their similarities in order to process them collaboratively. Our general estimation framework is based on the minimization of the quadratic risk, which is approximated in two steps, and adapts to photon and electronic noises. Relying on unbiased risk estimation (URE) for the first step and on ``internal adaptation'', a concept borrowed from deep learning theory, for the second, we show that our approach enables to reinterpret and reconcile previous state-of-the-art non-local methods. Within this framework, we propose a novel denoiser called NL-Ridge that exploits linear combinations of patches. While conceptually simpler, we show that NL-Ridge can outperform well-established state-of-the-art single-image denoisers.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper primarily focuses on addressing the problem of single image denoising, with a particular emphasis on non-local methods, and proposes a unified framework to reinterpret and reconcile existing state-of-the-art non-local denoising methods. Specifically, the paper addresses the following key issues:

1. **Unified Framework**: A general estimation framework based on quadratic risk minimization is proposed, which can reinterpret and integrate current advanced non-local denoising methods. The most notable representative of these methods is BM3D.

2. **New Method NL-Ridge**: Within this framework, a new denoiser called NL-Ridge is proposed, which utilizes a linear combination of similar patches for denoising. Despite its simple concept, experimental results show that NL-Ridge outperforms some existing state-of-the-art single image denoising methods.

3. **Addressing the Limitations of Supervised Learning**: While supervised learning methods are powerful, they rely on high-quality training datasets, which can be difficult to obtain in some cases (e.g., lack of noiseless images). The methods mentioned in the paper aim to overcome this limitation.

4. **Non-Local Strategy**: The methods in the paper are based on a non-local strategy, which denoises by collecting similar noisy patches in the image and processing them collaboratively. The core of this approach is to gather similar noisy patches together for collaborative denoising.

5. **Two-Step Method**: The method in the paper is divided into two steps: the first step uses Unbiased Risk Estimation (URE) to estimate weights; the second step employs an "internal adaptation" strategy to further optimize the weights.

6. **Different Types of Noise**: The paper considers different types of noise, including Gaussian noise, Poisson noise, and mixed Poisson-Gaussian noise, and demonstrates the applicability and effectiveness of its methods under these noise conditions.

In summary, the main goal of this research is to improve the denoising effect of single images by proposing a new, unified approach, especially when dealing with non-local patches. By introducing the NL-Ridge method, the authors not only provide a new denoising tool but also offer a new perspective for understanding existing state-of-the-art non-local denoising techniques.