Real-world Noisy Image Denoising: A New Benchmark

TL;DR

This paper targets real-world image denoising by creating a large, multi-camera benchmark capturing 40 scenes under six ISO levels, with ground-truth images obtained by averaging hundreds of frames. It evaluates a wide range of denoising methods on existing datasets and the new benchmark, showing that real-world noise is not AWGN and that methods tailored for real-world noise offer robustness on older datasets but face tighter margins on the new data. The results highlight the non-Gaussian, heteroscedastic, and cross-channel nature of real-world noise and establish a more challenging benchmark for advancing denoising research.

Abstract

Most of previous image denoising methods focus on additive white Gaussian noise (AWGN). However,the real-world noisy image denoising problem with the advancing of the computer vision techiniques. In order to promote the study on this problem while implementing the concurrent real-world image denoising datasets, we construct a new benchmark dataset which contains comprehensive real-world noisy images of different natural scenes. These images are captured by different cameras under different camera settings. We evaluate the different denoising methods on our new dataset as well as previous datasets. Extensive experimental results demonstrate that the recently proposed methods designed specifically for realistic noise removal based on sparse or low rank theories achieve better denoising performance and are more robust than other competing methods, and the newly proposed dataset is more challenging. The constructed dataset of real photographs is publicly available at \url{https://github.com/csjunxu/PolyUDataset} for researchers to investigate new real-world image denoising methods. We will add more analysis on the noise statistics in the real photographs of our new dataset in the next version of this article.

Figures (8)

• Figure 1: Captured images with the Sony A7 II camera under different (ISO, Shutter speed, Aperture) settings.
• Figure 2: The static scene is captured with a camera fixed by tripod. The data collection is automatically done with shutter release after the button is pressed by a person.
• Figure 3: Some sample images in our newly constructed dataset.
• Figure 4: Some cropped regions of the "ground truth" images (left) and their corresponding noisy images (right) in our constructed dataset.
• Figure 5: Denoised images and PSNR (dB)/SSIM results of the real-world noisy image Canon 5D Mark 3 ISO 3200 1crosschannel2016 by different methods. The images are better to be zoomed in on screen.