RENOIR - A Dataset for Real Low-Light Image Noise Reduction

Image denoising algorithms are evaluated using images corrupted by artificialnoise, which may lead to incorrect conclusions about their performances on realnoise. In this paper we introduce a dataset of color images corrupted bynatural noise due to low-light conditions, together with spatially andintensity-aligned low noise images of the same scenes. We also introduce amethod for estimating the true noise level in our images, since even the lownoise images contain small amounts of noise. We evaluate the accuracy of ournoise estimation method on real and artificial noise, and investigate thePoisson-Gaussian noise model. Finally, we use our dataset to evaluate sixdenoising algorithms: Active Random Field, BM3D, Bilevel-MRF, Multi-LayerPerceptron, and two versions of NL-means. We show that while the Multi-LayerPerceptron, Bilevel-MRF, and NL-means with soft threshold outperform BM3D ongray images with synthetic noise, they lag behind on our dataset.