Unprocessing Images for Learned Raw Denoising

TLDR

Machine‑learning denoising models perform best when trained on data that closely matches the evaluation domain, yet they are often trained on synthetic images while applied to real raw sensor readings, and the impact of post‑processing steps such as gain, color correction, and tone mapping is frequently ignored. We propose to unprocess images by inverting the image‑processing pipeline to synthesize realistic raw sensor measurements from Internet photos. The method models each processing component in the loss function, enabling training that accounts for all photometric transformations applied after denoising. Using this approach, a simple CNN achieves 14–38 % lower error rates and runs 9–18× faster than the prior state of the art on the Darmstadt Noise Dataset, while also generalizing to unseen sensors.

Abstract

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.