Learning Frequency Domain Priors for Image Demoireing
Learning Frequency Domain Priors for Image Demoireing
复制标题
DOI：
10.1109/tpami.2021.3115139
复制
发表时间：
2021-09
影响因子：
23.6
通讯作者：
Bolun Zheng;Shanxin Yuan;C. Yan;Xiang Tian;Jiyong Zhang;Yaoqi Sun;Lin Liu;A. Leonardis;G. Slaba
中科院分区：
文献类型：
--
作者：
Bolun Zheng;Shanxin Yuan;C. Yan;Xiang Tian;Jiyong Zhang;Yaoqi Sun;Lin Liu;A. Leonardis;G. Slaba
图像去噪是一个多方面的图像恢复任务，涉及到莫尔条纹去除和颜色恢复。本文提出了一种通用的退化模型来描述被莫尔条纹污染的图像，并提出了一种新的多尺度带通卷积神经网络（MBCNN）的单图像去噪。为了去除莫尔条纹，我们提出了一个多块大小的可学习带通滤波器（M-LBF），基于逐块频域变换，学习莫尔条纹的频域先验。我们还引入了一个新的损失函数命名为扩展高级Sobel损失（D-ASL），以更好地感知频率信息。对于颜色恢复，我们提出了两步色调映射策略，首先应用全局色调映射来校正全局色移，然后对每个像素的颜色进行局部微调。为了确定最合适的频域变换，我们研究了几种变换，包括DCT，DFT，DWT，可学习的非线性变换和可学习的正交变换。最后，我们采用了DCT。我们的基本模型赢得了AIM 2019演示挑战赛。在三个公共数据集上的实验结果表明，我们的方法比最先进的方法有很大的优势。
Image demoireing is a multi-faceted image restoration task involving both moire pattern removal and color restoration. In this paper, we raise a general degradation model to describe an image contaminated by moire patterns, and propose a novel multi-scale bandpass convolutional neural network (MBCNN) for single image demoireing. For moire pattern removal, we propose a multi-block-size learnable bandpass filters (M-LBFs), based on a block-wise frequency domain transform, to learn the frequency domain priors of moire patterns. We also introduce a new loss function named Dilated Advanced Sobel loss (D-ASL) to better sense the frequency information. For color restoration, we propose a two-step tone mapping strategy, which first applies a global tone mapping to correct for a global color shift, and then performs local fine tuning of the color per pixel. To determine the most appropriate frequency domain transform, we investigate several transforms including DCT, DFT, DWT, learnable non-linear transform and learnable orthogonal transform. We finally adopt the DCT. Our basic model won the AIM2019 demoireing challenge. Experimental results on three public datasets show that our method outperforms state-of-the-art methods by a large margin.