A generative adversarial network for image denoising

• Published in: Multimedia tools and applications Vol. 79; no. 23-24; pp. 16517 - 16529
• Main Authors: Zhong, Yue, Liu, Lizhuang, Zhao, Dan, Li, Hongyang
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2020; Springer Nature B.V
• Subjects: Artificial neural networks; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Generative adversarial networks; Image quality; Multimedia; Multimedia Information Systems; Neural networks; Noise; Noise reduction; Special Purpose and Application-Based Systems; Densenet; Generative adversarial network; Wasserstein-GAN; Image denoising
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-019-7556-x

Recent studies have shown that the performance of image denoising methods can be improved significantly by using deep convolutional neural networks(CNN). The traditional CNN ways mainly focus on minimizing the Mean Squared Error (MSE), resulting in a feeling that the images lack of high-frequency details. So we apply a generative adversarial network (GAN) in image denoising. A very deep convolutional densenet framework is acting as our generator, which benefits in easing the vanishing-gradient problem of very deep networks. Moreover, we use Wasserstein-GAN as our loss function to stabilize the training process. Also, the Wasserstein distance between real and generated images from discriminator can be regarded as an indicator that has been proved highly relevant to the quality of the generated sample. A photo-realistic image with higher quality can be produced through our work than in traditional ways.