Benchmarking Denoising Algorithms with Real Photographs

Lacking realistic ground truth data, image denoising techniques are traditionally evaluated on images corrupted by synthesized i.i.d. Gaussian noise. We aim to obviate this unrealistic setting by developing a methodology for benchmarking denoising techniques on real photographs. We capture pairs of images with different ISO values and appropriately adjusted exposure times, where the nearly noise-free low-ISO image serves as reference. To derive the ground truth, careful post-processing is needed. We correct spatial misalignment, cope with inaccuracies in the exposure parameters through a linear intensity transform based on a novel heteroscedastic Tobit regression model, and remove residual low-frequency bias that stems, e.g., from minor illumination changes. We then capture a novel benchmark dataset, the Darmstadt Noise Dataset (DND), with consumer cameras of differing sensor sizes. One interesting finding is that various recent techniques that perform well on synthetic noise are clearly outperformed by BM3D on photographs with real noise. Our benchmark delineates realistic evaluation scenarios that deviate strongly from those commonly used in the scientific literature.

• Publication year

  2017

• Venue

  Computer Vision and Pattern Recognition

• Publication date

  2017-07-05

• Fields of study

  Computer Science

• Identifiers
  DOI 10.1109/CVPR.2017.294arXiv 1707.01313
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• The Darmstadt Noise Dataset (DND) provides a new benchmark for real photographs, and BM3D is reported to outperform several recent techniques that perform well under synthetic i.i.d. Gaussian noise.

  Confidence 0.94

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• Ground-truth derivation corrects spatial misalignment, fits a linear intensity transform with a heteroscedastic Tobit regression model, and removes residual low-frequency bias.

  Confidence 0.95

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• Real photographs are acquired as pairs at different ISO values and exposure times, with the nearly noise-free low-ISO reference image serving as the ground-truth reference after post-processing.

  Confidence 0.97

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review

• bm3d

  algorithm, baseline

  A classical denoising algorithm used as a baseline comparator on the benchmark.

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• darmstadt noise dataset (dnd)

  dataset, benchmark

  The benchmark dataset of real noisy photographs collected for denoising evaluation.

  Aliases: DND

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• exposure times

  camera setting, parameter

  Shutter durations adjusted alongside ISO to balance brightness between paired captures.

  Aliases: exposure time, shutter time

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• heteroscedastic tobit regression model

  statistical model, method

  A regression model used to estimate a linear intensity transform while accounting for heteroscedastic censored observations in the paired images.

  Aliases: Tobit regression model, heteroscedastic Tobit

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• iso values

  camera setting, parameter

  Camera sensitivity settings used to acquire paired images at different noise levels.

  Aliases: ISO, ISO setting

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• low-iso reference image

  reference image, data

  The nearly noise-free low-ISO image in each pair that serves as the reference for ground-truth derivation.

  Aliases: reference image, low-ISO image

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• real photographs

  evaluation setting, data

  Photographic images captured from real scenes and used here as the evaluation domain instead of synthetic corruption.

  Aliases: real photo, photographs

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review
• synthetic i.i.d. gaussian noise

  noise model, evaluation setting

  Artificial independently and identically distributed Gaussian corruption commonly used in traditional denoising evaluation.

  Aliases: synthesized i.i.d. Gaussian noise, Gaussian noise

  박진우 (dztg5apj7m) extractionB (s683577b42) review--------- ✂ Cut Here ✂ --------- (jqthcshryb) review

• From Noise Modeling to Blind Image Denoising

  2016cited by this paper
• A Software Platform for Manipulating the Camera Imaging Pipeline

  2016cited by this paper
• A Holistic Approach to Cross-Channel Image Noise Modeling and Its Application to Image Denoising

  2016cited by this paper
• Quantile analysis of image sensor noise distribution

  2015cited by this paper
• On learning optimized reaction diffusion processes for effective image restoration

  2015cited by this paper
• Shrinkage Fields for Effective Image Restoration

  2014cited by this paper
• Noise Parameter Mismatch in Variance Stabilization, With an Application to Poisson–Gaussian Noise Estimation

  2014cited by this paper
• RENOIR - A dataset for real low-light image noise reduction

  2014influential reference
• Weighted Nuclear Norm Minimization with Application to Image Denoising

  2014influential reference
• Gaussian-Cauchy mixture modeling for robust signal-dependent noise estimation

  2014cited by this paper
• Practical Signal-Dependent Noise Parameter Estimation From a Single Noisy Image

  2014cited by this paper
• Nonlocally Centralized Sparse Representation for Image Restoration

  2013influential reference
• Optimal Inversion of the Generalized Anscombe Transformation for Poisson-Gaussian Noise

  2013cited by this paper
• Image denoising: Can plain neural networks compete with BM3D?

  2012influential reference
• How Well Do Filter-Based MRFs Model Natural Images?

  2012cited by this paper
• Loss-Specific Training of Non-Parametric Image Restoration Models: A New State of the Art

  2012cited by this paper
• From learning models of natural image patches to whole image restoration

  2011cited by this paper
• Image Denoising in Mixed Poisson–Gaussian Noise

  2011cited by this paper
• EMVA Standard 1288 Standard for Characterization of Image Sensors and Cameras

  2010cited by this paper
• Guided Image Filtering

  2010cited by this paper
• Non-local sparse models for image restoration

  2009cited by this paper
• A case for denoising before demosaicking color filter array data

  2009cited by this paper
• Clipped noisy images: Heteroskedastic modeling and practical denoising

  2009influential reference
• Fields of Experts

  2009cited by this paper
• Efficient subpixel image registration algorithms.

  2008cited by this paper
• Practical Poissonian-Gaussian Noise Modeling and Fitting for Single-Image Raw-Data

  2008influential reference
• Noise Measurement for Raw-Data of Digital Imaging Sensors by Automatic Segmentation of Nonuniform Targets

  2007cited by this paper
• Image Denoising by Sparse 3-D Transform-Domain Collaborative Filtering

  2007influential reference
• Denoising Archival Films using a Learned Bayesian Model

  2006cited by this paper
• $rm K$-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse Representation

  2006cited by this paper
• CMOS image sensors

  2005cited by this paper
• K-SVD : An Algorithm for Designing of Overcomplete Dictionaries for Sparse Representation

  2005cited by this paper
• Image denoising by non-local averaging

  2005cited by this paper
• Image quality assessment: from error visibility to structural similarity

  2004cited by this paper
• Image denoising using scale mixtures of Gaussians in the wavelet domain

  2003cited by this paper
• Image processing and data analysis: the multiscale approach

  1998cited by this paper
• Radiometric CCD camera calibration and noise estimation

  1994cited by this paper
• An Iterative Image Registration Technique with an Application to Stereo Vision

  1981cited by this paper
• Estimation of Relationships for Limited Dependent Variables

  1958influential reference
• Submitted to Ieee Transactions on Pattern Analysis and Machine Intelligence Automatic Estimation and Removal of Noise from a Single Image

  year unknowncited by this paper

• Progressive feature refinement network for blind gaussian image denoising

  2026cites this paper
• A Practical CNN–Transformer Hybrid Network for Real-World Image Denoising

  2026cites this paper
• Nonlocal multiplicative denoising method based on variational approach and topological shape optimization

  2026cites this paper
• Teacher-Guided Causal Interventions for Image Denoising: Orthogonal Content-Noise Disentanglement in Vision Transformers

  2026cites this paper
• Dark3R: Learning Structure from Motion in the Dark

  2026cites this paper
• Enhancing self-supervised image denoising with asymmetric mask blind-spot network

  2026cites this paper
• Diffusion-Based sRGB Real Noise Generation via Prompt-Driven Noise Representation Learning

  2026cites this paper
• MSCNRNet: Non-dense residual network with multi-skip connection for image restoration

  2026cites this paper
• RoGER-SLAM: A Robust Gaussian Splatting SLAM System for Noisy and Low-Light Environment Resilience

  2025cites this paper
• DADU-Net: Degradation-Aware Dual U-Net with Soft Routing and Fusion for Single-Image Blind Restoration

  2025cites this paper
• An efficient U-shaped transformer network for low-light power image denoising

  2025cites this paper
• DMRNet: Dual Multi-Granularity Residual Attention Network Towards Real Image Denoising

  2025cites this paper
• SNIC: Synthesized Noisy Images using Calibration

  2025cites this paper
• Learning-Based Compression for Noisy Images in the Wild

  2025cites this paper
• A Review of Neural Network-Based Image Noise Processing Methods

  2025influential citation
• DarkEQA: Benchmarking Vision-Language Models for Embodied Question Answering in Low-Light Indoor Environments

  2025cites this paper
• Model Extraction for Image Denoising Networks

  2025cites this paper
• Camera Sensor Raw Data-Driven Video Blur Effect Prevention: Dataset and Study

  2025cites this paper
• Learnable mixture distribution prior for image denoising

  2025cites this paper
• Net2Net: When Un-trained Meets Pre-trained Networks for Robust Real-World Denoising

  2025cites this paper
• Self-Supervised Real-World Denoising with Mixed Pixel Shuffle and Blind-Spot Network

  2025cites this paper
• Image Denoising Using Global and Local Circulant Representation

  2025cites this paper
• A Parallel Image Denoising Network Based on Nonparametric Attention and Multiscale Feature Fusion

  2025cites this paper
• DeepSN-Net: Deep Semi-Smooth Newton Driven Network for Blind Image Restoration

  2025cites this paper
• Masked Video Pretraining Advances Real-World Video Denoising

  2025cites this paper
• Remote sensing image Super-resolution reconstruction by fusing multi-scale receptive fields and hybrid transformer

  2025cites this paper
• Multispectral Demosaicing via Dual Cameras

  2025cites this paper
• Dual Path Unsupervised Real Image Denoising

  2025cites this paper
• Unsupervised Real-World Denoising: Sparsity is All You Need

  2025cites this paper
• Grouped spatial attentional network with regularization for real world image denoising

  2025cites this paper
• Dual-Path Adversarial Denoising Network Based on UNet

  2025cites this paper
• Effective enhancement and fusion of multi-perspective features for self-supervised real image denoising

  2025cites this paper
• Two-stage Mamba-based diffusion model for image restoration

  2025cites this paper
• Efficient Image Denoising Using Global and Local Circulant Representation

  2025cites this paper
• Supervised Denoising for Extreme Low-Light Raw Videos

  2025cites this paper
• ColorFormer: A novel colorization method based on a transformer

  2025cites this paper
• Physical Degradation Model-Guided Interferometric Hyperspectral Reconstruction with Unfolding Transformer

  2025cites this paper
• Beyond Image Prior: Embedding Noise Prior into Latent Space of Conditional Denoising Transformer

  2025cites this paper
• CellINR: Implicitly Overcoming Photo-induced Artifacts in 4D Live Fluorescence Microscopy

  2025cites this paper
• New Trends in Image Restoration based on Artificial Intelligent Models: Analytical Study

  2025cites this paper
• WA‐BSN: Self‐Supervised Real‐World Image Denoising Based on Wavelet‐Adaptive Blind Spot Network

  2025influential citation
• Self-supervised multi-blind network for real image denoising via multivariate Gaussian-poisson noise

  2025cites this paper
• DichotomyIR: Universal Image Reconstruction via Dichotomy Classification and Uncertainty Elimination

  2025cites this paper
• UnfoldLDM: Deep Unfolding-based Blind Image Restoration with Latent Diffusion Priors

  2025cites this paper
• CARD: Correlation Aware Restoration with Diffusion

  2025cites this paper
• ZERO-PDE: Zero-shot joint denoising and enhancement for infrared polarization images

  2025cites this paper
• DMANet: An Image Denoising Network Based on Dual Convolutional Neural Networks with Multiple Attention Mechanisms

  2025cites this paper
• Learning Cocoercive Conservative Denoisers via Helmholtz Decomposition for Poisson Inverse Problems

  2025cites this paper
• Rotation-Equivariant Self-Supervised Method in Image Denoising

  2025cites this paper
• Overview of Research on Digital Image Denoising Methods

  2025cites this paper
• Design Principles of Multi-Scale $\mathcal{J}$-Invariant Networks for Self-Supervised Image Denoising

  2025cites this paper
• VibrantLeaves: A principled parametric image generator for training deep restoration models

  2025cites this paper
• Implicit Multi-Scale Swin Transformer Network for Image Denoising

  2025cites this paper
• A Graph-Based Accelerator of Retinex Model With Bit-Serial Computing for Image Enhancements

  2025cites this paper
• Leveraging Frequency Analysis for Image Denoising Network Pruning

  2025cites this paper
• LLFA: Fusing Global Illumination and Local Priors for Low-Light Face Image Enhancement with Adaptor

  2025cites this paper
• 基于记忆单元的多阶段图像自监督去噪方法

  2025cites this paper
• Dream-IF: Dynamic Relative EnhAnceMent for Image Fusion

  2025cites this paper
• Calibration-Free Raw Image Denoising via Fine-Grained Noise Estimation

  2025influential citation
• RestorMamba: An Enhanced Synergistic State Space Model for Image Restoration

  2025cites this paper
• Visual and large multimodal models promote image restoration and enhancement： research progress

  2025cites this paper
• CAWSNet: Channel attention weight sharing network for single image denoising

  2025cites this paper
• AIM 2025 Low-Light RAW Video Denoising Challenge: Dataset, Methods and Results

  2025cites this paper
• Examining Joint Demosaicing and Denoising for Single-, Quad-, and Nona-Bayer Patterns

  2025cites this paper
• A Mamba U-Net Model for Reconstruction of Extremely Dark RGGB Images

  2025cites this paper
• Natural-light full-color motion-picture holography

  2025cites this paper
• PFCPNet: A progressive feature correction and prompt network for robust real-world image denoising

  2025cites this paper
• A Tightly Coupled AI-ISP Vision Processor

  2025cites this paper
• ECA-MSNet: A Multi-Scale Residual U-Net with Efficient Channel Attention for Real-World Image Denoising

  2025cites this paper
• Deep Learning for Computational Imaging

  2025cites this paper
• ReCalKV: Low-Rank KV Cache Compression via Head Reordering and Offline Calibration

  2025cites this paper
• YOND: Practical Blind Raw Image Denoising Free from Camera-Specific Data Dependency

  2025cites this paper
• Super Resolution Image Denoiser Network: Toward Real-world Image Noise Removal

  2025influential citation
• Learning Multi-scale Spatial-frequency Features for Image Denoising

  2025influential citation
• Deep Learning-Based Image Denoising: A Comprehensive Survey

  2025cites this paper
• ASTSFormer: adaptive sparse token selective transformer with superpixel-guide for low-light image denoising

  2025cites this paper
• Pureformer: Transformer-Based Image Denoising

  2025cites this paper
• P‐5.19: Detecting Deepfake Images by Noisy Pattern

  2025cites this paper
• A Review of Optical Image Enhancement for Extreme Space Environments

  2025cites this paper
• RDDM: Practicing RAW Domain Diffusion Model for Real-world Image Restoration

  2025cites this paper
• DarkVRAI: Capture-Condition Conditioning and Burst-Order Selective Scan for Low-light RAW Video Denoising

  2025cites this paper
• Image Denoising Meets Quantization: Exploring the Effects of Post-Training Quantization

  2025cites this paper
• SERDNet: adaptive residual dense learning with dual-path fusion and squeeze-and-excitation for robust image denoising

  2025cites this paper
• Blind-Spot Guided Diffusion for Self-supervised Real-World Denoising

  2025cites this paper
• Domain-connected gradient-enhanced transformer network for robust image denoising

  2025cites this paper
• Two-stage Progressive Residual Dense Attention Network for Image Denoising

  2024cites this paper
• Toward Robust Image Denoising via Flow-Based Joint Image and Noise Model

  2024cites this paper
• Masked Pre-training Enables Universal Zero-shot Denoiser

  2024cites this paper
• CascadedGaze: Efficiency in Global Context Extraction for Image Restoration

  2024cites this paper
• DA-BSN: Self-Supervised Real-World Image Denoising based on Dynamic Adaptive Blind Spot Network

  2024cites this paper
• Multiscale Synergism Ensemble Progressive and Contrastive Investigation for Image Restoration

  2024cites this paper
• Masked Pre-trained Model Enables Universal Zero-shot Denoiser

  2024cites this paper
• AGP-Net: Adaptive Graph Prior Network for Image Denoising

  2024cites this paper
• How to Best Combine Demosaicing and Denoising?

  2024cites this paper
• An Image Denoising Algorithm Based on Multi-scale Residual Channel Attention Network

  2024influential citation
• Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising

  2024cites this paper
• Dual-GAN Complementary Learning for Real-World Image Denoising

  2024cites this paper
• Self-Supervised Real-World Denoising by Jointly Learning Visible and Invisible Noise

  2024cites this paper
• Efficient denoising method for real-world noise image using Scalable Convolution and Channel Interaction Attention

  2024cites this paper
• Joint multi-dimensional dynamic attention and transformer for general image restoration

  2024cites this paper