Image denoising using convolutional neural networks and the Natural Image Noise Dataset

Convolutional neural networks have been the focus of research aiming to solve image denoising problems but their performance remains unsatisfactory for most applications. These networks are trained with synthetic noise distributions that do not accurately reflect the noise captured by image sensors. Some datasets of clean-noisy image pairs have been introduced but they are usually meant for benchmarking or specific applications. We introduce the Natural Image Noise Dataset (NIND), a dataset of DSLR-like images with varying levels of ISO noise that is large enough to train models for blind denoising over a wide range of noise. We demonstrate the use of denoising models trained with the NIND and show that they significantly outperform BM3D on ISO noise from unseen images, even when generalizing to images from a different type of camera. We expect that this dataset will prove useful for future image denoising applications. In support of this, we investigate the use of conditional generative adversarial networks (cGANs), which provide better denoising performance in some edge cases that a paired dataset may not cover. This has enabled us to develop a framework that lays the groundwork for training (c)GANs for image denoising purposes, as well as introduce a pair of discriminator-generator architectures that perform well in such systems. This work provides proof of concept support for the use of deep learning for photographic image denoising, as facilitated by the NIND.