Triple discriminators - equipped GAN for Denoising of Chinese calligraphic tablet images

Denoising of Chinese calligraphic tablet images is of great importance in regard to the study of both content and character shapes in these images. Formerly GAN (generative adversarial network) based image denoising methods model the noise in the generator and then perform denoising by CNN (convolutional neural networks) algorithms. These methods still leave room for improvement. In this paper, a triple discriminators equipped GAN for generative denoising is proposed, with the three channels of discriminators enhancing the denoising result by different means. Another noise modeling module based on CycleGAN is used to produce the paired input data. Quantitative index are obtained for these methods; the PSNR and SSIM of our method on publicly available data is 21.84 and 0.93 respectively, which is preferable to BM3D, DnCNN, FormResNet, CycleGAN and our previous method.