BDGAN: Image Blind Denoising Using Generative Adversarial Networks

In this paper, we present an end-to-end method for image blind denoising based on a conditional generative adversarial network (GAN). Discriminative learning-based methods, such as DnCNN, can achieve state-of-the-art denoising results but these methods usually focus on establishing noise model that resembles natural noisy images, thus neglecting to recover clean images from noisy images. Non-blind denoising methods are also limited since a precise noise level is hard to be obtained in the real world. Using multiple modified methods, we propose a novel end-to-end architecture which could directly generate clean images. A range of experiments have been done to show the convenience and superiority of our approach in image blind denoising.