[论文解读] Universal Denoising Networks : A Novel CNN-based Network Architecture for Image Denoising.

Stamatios Lefkimmiatis|arXiv (Cornell University)|Nov 21, 2017

Image and Signal Denoising Methods被引用 11

一句话总结

该论文提出两种基于CNN的新型去噪网络——一种使用卷积层，另一种使用非局部滤波层——联合利用自然图像中的局部和非局部自相似性。网络经过一次训练即可处理广泛的噪声水平，在加性白高斯噪声下实现最先进性能，参数量仅为现有模型的1/10，同时对噪声统计不匹配保持鲁棒性。

ABSTRACT

We design a novel network architecture for learning discriminative image models that are employed to efficiently tackle the problem of grayscale and color image denoising. Based on the proposed architecture, we introduce two different variants. The first network involves convolutional layers as a core component, while the second one relies instead on non-local filtering layers and thus it is able to exploit the inherent non-local self-similarity property of natural images. As opposed to most of the existing neural networks, which require the training of a specific model for each considered noise level, the proposed networks are able to handle a wide range of different noise levels, while they are very robust when the noise degrading the latent image does not match the statistics of the one used during training. The latter argument is supported by results that we report on publicly available images corrupted by unknown noise and which we compare against solutions obtained by alternative state-of-the-art methods. At the same time the introduced networks achieve excellent results under additive white Gaussian noise (AWGN), which are comparable to the current state-of-the-art network, while they depend on a more shallow architecture with the number of trained parameters being one order of magnitude smaller. These properties make the proposed networks ideal candidates to serve as sub-solvers on restoration methods that deal with general inverse imaging problems such as deblurring, demosaicking, superresolution, etc.

研究动机与目标

• 开发单一深度学习模型，无需微调即可泛化至多种噪声水平。
• 利用自然图像中的非局部自相似性以提升去噪性能。
• 在保持或超越现有最先进方法性能的同时，降低模型复杂度。
• 为更广泛的逆成像问题（如去模糊和超分辨率）创建有效的子求解器。

提出的方法

• 第一种变体使用标准卷积层提取局部特征，并从噪声输入重建干净图像。
• 第二种变体用非局部滤波层替代卷积层，以利用图像中的长程依赖关系和自相似结构。
• 两种架构均在多样化噪声水平上端到端训练，以实现泛化能力。
• 网络设计为浅层但高效，最大限度减少可训练参数数量。
• 架构优化以处理训练期间未见的噪声统计，增强鲁棒性。
• 模型在合成图像和具有未知噪声特征的真实世界图像上进行评估。

实验结果

研究问题

• RQ1 能否训练单一深度神经网络，在无需微调的情况下对广泛噪声水平的图像进行去噪？
• RQ2 与标准卷积层相比，引入非局部滤波层在图像去噪性能上有多大提升？
• RQ3 在加性白高斯噪声下，参数更少的轻量化网络在性能上能与最先进模型达到何种可比程度？
• RQ4 当测试图像受到与训练分布不匹配的噪声污染时，所提出的网络有多强的鲁棒性？
• RQ5 此类去噪网络能否作为更广泛逆成像流程中的有效子求解器？

主要发现

• 所提出的网络在加性白高斯噪声下实现最先进性能，与最佳现有方法相当。
• 相比当前最先进网络，模型参数量仅需其十分之一，显著降低模型复杂度。
• 网络在各种噪声水平下泛化良好，即使测试噪声与训练噪声统计不同亦表现优异。
• 非局部变体通过更有效地利用图像自相似性，优于卷积变体。
• 在具有未知噪声的真实世界图像上，模型表现出强鲁棒性，优于其他最先进方法。
• 由于其良好的泛化能力和高效性，该架构非常适合作为去模糊和超分辨率等逆成像流程中的子求解器。