Date:2016
Author:Kai Zhang, Wangmeng Zuo, Yunjin Chen, Deyu Meng, and Lei Zhang
Pages:13
Summary:This paper introduces a deep convolutional neural network (DnCNN) for image denoising, focusing on the integration of residual learning and batch normalization to enhance training efficiency and denoising performance. Unlike traditional methods that train specific models for fixed noise levels, DnCNN is designed to handle blind Gaussian denoising with unknown noise levels. The network predicts the residual image, which is the difference between the noisy observation and the latent clean image, by removing the clean image from the noisy observation in the hidden layers. Batch normalization is used to stabilize and speed up the training process. Extensive experiments demonstrate that DnCNN outperforms state-of-the-art methods in terms of both quantitative metrics and visual quality, and it can be efficiently implemented using GPU computing. Additionally, the paper shows that a single DnCNN model can be trained to handle multiple general image denoising tasks, including blind Gaussian denoising, single image super-resolution, and JPEG image deblocking.