Unprocessing Images for Learned Raw Denoising

Machine learning techniques work best when the data used for training resembles the data used for evaluation. This holds true for learned single-image denoising algorithms, which are applied to real raw camera sensor readings but, due to practical constraints, are often trained on synthetic image data. Though it is understood that generalizing from synthetic to real images requires careful consideration of the noise properties of camera sensors, the other aspects of an image processing pipeline (such as gain, color correction, and tone mapping) are often overlooked, despite their significant effect on how raw measurements are transformed into finished images. To address this, we present a technique to “unprocess” images by inverting each step of an image processing pipeline, thereby allowing us to synthesize realistic raw sensor measurements from commonly available Internet photos. We additionally model the relevant components of an image processing pipeline when evaluating our loss function, which allows training to be aware of all relevant photometric processing that will occur after denoising. By unprocessing and processing training data and model outputs in this way, we are able to train a simple convolutional neural network that has 14%-38% lower error rates and is 9×-18× faster than the previous state of the art on the Darmstadt Noise Dataset, and generalizes to sensors outside of that dataset as well.

1. Introduction

Traditional single-image denoising algorithms often analytically model properties of images and the noise they are designed to remove. In contrast, modern denoising methods often employ neural networks to learn a mapping from noisy images to noise-free images. Deep learning is capable of representing complex properties of images and noise, but training these models requires large paired datasets. As a result, most learning-based denoising techniques rely on synthetic training data. Despite significant work on designing neural networks for denoising, recent benchmarks [3], [31] reveal that deep learning models are often outperformed by traditional, hand-engineered algorithms when evaluated on real noisy raw images.