Noise2Score: Tweedie's Approach to Self-Supervised Image Denoising\n without Clean Images

2021/06/13 by Kwanyoung Kim, Jong Chul Ye, Kim, Kwanyoung +1 · 20 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Earth and Planetary Sciences · Engineering · #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image Processing Techniques and Applications #Image and Signal Denoising Methods #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Seismic Imaging and Inversion Techniques #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.2106.07009

openalex publication_date 2021/06/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Recently, there has been extensive research interest in training deep\nnetworks to denoise images without clean reference. However, the representative\napproaches such as Noise2Noise, Noise2Void, Stein's unbiased risk estimator\n(SURE), etc. seem to differ from one another and it is difficult to find the\ncoherent mathematical structure. To address this, here we present a novel\napproach, called Noise2Score, which reveals a missing link in order to unite\nthese seemingly different approaches. Specifically, we show that image\ndenoising problems without clean images can be addressed by finding the mode of\nthe posterior distribution and that the Tweedie's formula offers an explicit\nsolution through the score function (i.e. the gradient of log likelihood). Our\nmethod then uses the recent finding that the score function can be stably\nestimated from the noisy images using the amortized residual denoising\nautoencoder, the method of which is closely related to Noise2Noise or\nNose2Void. Our Noise2Score approach is so universal that the same network\ntraining can be used to remove noises from images that are corrupted by any\nexponential family distributions and noise parameters. Using extensive\nexperiments with Gaussian, Poisson, and Gamma noises, we show that Noise2Score\nsignificantly outperforms the state-of-the-art self-supervised denoising\nmethods in the benchmark data set such as (C)BSD68, Set12, and Kodak, etc.\n

Citations

• Tweedie’s Formula and Selection Bias
• Estimation of the Mean of a Multivariate Normal Distribution
• Noise2Same: Optimizing A Self-Supervised Bound for Image Denoising

Cited by

• Restoration Flow Matching-Based Channel Refinement and Equalization Correction for MIMO Semantic Communications
• Self-Supervised Learning from Noisy and Incomplete Data
• Diffusion Posterior Sampler for Hyperspectral Unmixing with Spectral Variability Modeling
• Predicting partially observable dynamical systems via diffusion models with a multiscale inference scheme
• cryoSENSE: Compressive Sensing Enables High-throughput Microscopy with Sparse and Generative Priors on the Protein Cryo-EM Image Manifold
• Equivariant Sampling for Improving Diffusion Model-based Image Restoration
• Toward the Frontiers of Reliable Diffusion Sampling via Adversarial Sinkhorn Attention Guidance
• Text to Sketch Generation with Multi-Styles
• Adversarial and Score-Based CT Denoising: CycleGAN vs Noise2Score
• LumaGuide: Distribution Shaping for Training-Free HDR Generation in Diffusion Models
• Moments Matter: Posterior Recovery in Poisson Denoising via Log-Networks
• Tunable-Generalization Diffusion Powered by Self-Supervised Contextual Sub-Data for Low-Dose CT Reconstruction
• SimDiff: Simulator-constrained Diffusion Model for Physically Plausible Motion Generation
• Incomplete Data, Complete Dynamics: A Diffusion Approach
• Single-Image Depth from Defocus with Coded Aperture and Diffusion Posterior Sampling
• Generalizable Holographic Reconstruction via Amplitude-Only Diffusion Priors
• A Survey of Deep Learning-based Point Cloud Denoising
• Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration
• Diffusion models for inverse problems
• Robust multi‐coil MRI reconstruction via self‐supervised denoising

Related

• Noise Distribution Adaptive Self-Supervised Image Denoising using Tweedie Distribution and Score Matching
• Noise2Same: Optimizing A Self-Supervised Bound for Image Denoising
• Noise2Blur: Online Noise Extraction and Denoising
• Self2Self+: Single-Image Denoising with Self-Supervised Learning and Image Quality Assessment Loss
• Noise2Inpaint: Learning Referenceless Denoising by Inpainting Unrolling