Noise2Self: Blind Denoising by Self-Supervision

2019/01/30 by Joshua Batson, Löıc A. Royer, Batson, Joshua +2 · 49 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Engineering · Mathematics · #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Processing Techniques and Applications #Image and Signal Denoising Methods #Machine Learning (cs.LG) #Machine Learning (stat.ML) #cs.CV #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.1901.11365

10 pages, 6 figures, and supplement

openalex publication_date 2019/01/30 · arxiv created 2019/06/08 · arxiv updated 2019/06/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We propose a general framework for denoising high-dimensional measurements which requires no prior on the signal, no estimate of the noise, and no clean training data. The only assumption is that the noise exhibits statistical independence across different dimensions of the measurement, while the true signal exhibits some correlation. For a broad class of functions ("J-invariant"), it is then possible to estimate the performance of a denoiser from noisy data alone. This allows us to calibrate J-invariant versions of any parameterised denoising algorithm, from the single hyperparameter of a median filter to the millions of weights of a deep neural network. We demonstrate this on natural image and microscopy data, where we exploit noise independence between pixels, and on single-cell gene expression data, where we exploit independence between detections of individual molecules. This framework generalizes recent work on training neural nets from noisy images and on cross-validation for matrix factorization.

Cited by

• Self-Supervised Learning from Noisy and Incomplete Data
• Next-Scale Prediction: A Self-Supervised Approach for Real-World Image Denoising
• SFBD-OMNI: Bridge models for lossy measurement restoration with limited clean samples
• DNA-Prior: Unsupervised Denoise Anything via Dual-Domain Prior
• An unsupervised posterior sampling framework for multi-purpose seismic data recovery
• Equivariant Sampling for Improving Diffusion Model-based Image Restoration
• Locally-Supervised Global Image Restoration
• Equivariance2Inverse: A Practical Self-Supervised CT Reconstruction Method Benchmarked on Real, Limited-Angle, and Blurred Data
• Lightweight Data-Free Denoising for Detail-Preserving Biomedical Image Restoration
• SDPA++: A General Framework for Self-Supervised Denoising with Patch Aggregation
• PC-UNet: An Enforcing Poisson Statistics U-Net for Positron Emission Tomography Denoising
• Median2Median: Zero-shot Suppression of Structured Noise in Images
• Adaptive Guided Upsampling for Low-light Image Enhancement
• Measuring and Controlling the Spectral Bias for Self-Supervised Image Denoising
• Tunable-Generalization Diffusion Powered by Self-Supervised Contextual Sub-Data for Low-Dose CT Reconstruction
• Joint Demosaicking and Denoising in the Wild: The Case of Training Under Ground Truth Uncertainty
• Blind-Spot Guided Diffusion for Self-supervised Real-World Denoising
• Improving Accuracy and Efficiency of Implicit Neural Representations: Making SIREN a WINNER
• Spectral Bottleneck in Sinusoidal Representation Networks: Noise is All You Need
• A Graph Laplacian Eigenvector-based Pre-training Method for Graph Neural Networks
• IDF: Iterative Dynamic Filtering Networks for Generalizable Image Denoising
• CellINR: Implicitly Overcoming Photo-induced Artifacts in 4D Live Fluorescence Microscopy
• Backdooring Self-Supervised Contrastive Learning by Noisy Alignment
• Learning Model-Blind Temporal Denoisers without Ground Truths
• Efficient Image Denoising Using Global and Local Circulant Representation
• Noise Robust Generative Adversarial Networks
• NoiseSDF2NoiseSDF: Learning Clean Neural Fields from Noisy Supervision
• Legacy Photo Editing with Learned Noise Prior
• Speckle2Self: Self-Supervised Ultrasound Speckle Reduction Without Clean Data
• Self-supervised Deep Learning for Denoising in Ultrasound Microvascular Imaging
• Lightweight Physics-Aware Zero-Shot Ultrasound Plane-Wave Denoising
• Zero-shot Denoising via Neural Compression: Theoretical and algorithmic framework
• Restore from Restored: Single-image Inpainting
• W2S: Microscopy Data with Joint Denoising and Super-Resolution for Widefield to SIM Mapping
• Image Restoration Learning via Noisy Supervision in the Fourier Domain
• Is Noise Conditioning Necessary? A Unified Theory of Unconditional Graph Diffusion Models
• Learning Invariant Representation for Unsupervised Image Restoration
• Deep Spectral Prior
• Rotation-Equivariant Self-Supervised Method in Image Denoising
• DeepInverse: A Python package for solving imaging inverse problems with deep learning
• Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum
• Nash: Neural Adaptive Shrinkage for Structured High-Dimensional Regression
• Noise-Robust Conditional Flow Matching: Generating Clean Samples from Noisy Datasets
• NeRF in the Dark: High Dynamic Range View Synthesis from Noisy Raw Images
• Restore from Restored: Single Image Denoising with Pseudo Clean Image
• Image Deconvolution via Noise-Tolerant Self-Supervised Inversion
• Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising
• Self-Supervised Noise Adaptive MRI Denoising via Repetition to Repetition (Rep2Rep) Learning
• Filter2Noise: Interpretable Self-Supervised Single-Image Denoising for Low-Dose CT with Attention-Guided Bilateral Filtering
• V2V3D: View-to-View Denoised 3D Reconstruction for Light-Field Microscopy

Related

• Joint self-supervised blind denoising and noise estimation
• Noise2Same: Optimizing A Self-Supervised Bound for Image Denoising
• Self-supervised denoising for massive noisy images
• Noise2Kernel: Adaptive Self-Supervised Blind Denoising using a Dilated Convolutional Kernel Architecture
• Noise2Score: Tweedie's Approach to Self-Supervised Image Denoising without Clean Images