Research map / Image generation / Image restoration
Benchmarking Denoising Algorithms With Real Photographs
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply variance-stabilizing transformation before BM3D denoising (ablation) | BM3D on DND RAW images · DND, RAW → RAW | PSNR (dB): 46.63 -> 47.14 | Helped |
| Apply variance-stabilizing transformation before WNNM denoising (ablation) | WNNM on DND RAW images · DND, RAW → RAW | PSNR (dB): 46.29 -> 47.10 | Helped |
| Apply variance-stabilizing transformation before FoE denoising (ablation) | FoE on DND RAW images · DND, RAW → RAW | PSNR (dB): 45.77 -> 44.11 | Hurt |
| Denoise linear RAW images rather than converted sRGB images (ablation) | DND denoising algorithms · DND benchmark | PSNR | Helped |
| Use BM3D instead of recent methods that were favored on synthetic Gaussian noise | DND real-photograph denoising · DND benchmark | PSNR | Helped |
| Use the full post-processing pipeline rather than no post-processing (ablation) | Low-ISO reference image ground truth · Simulated image pairs | PSNR: 43.14 -> 53.71 | Helped |
| Add heteroscedastic noise modeling to the Tobit regression (ablation) | Linear intensity scaling estimation · Simulated image pairs | RMSE of intensity-scaling estimates | Helped |
| Model clipped observations and intensity-dependent noise in the regression (ablation) | Ground-truth intensity alignment · Real and simulated image pairs | Linear correlation of residual with debiased image: 0.1874 -> 0.0051 | Helped |
| Remove low-frequency residuals in the debiased domain (ablation) | Reference-image ground-truth construction · Simulated image pairs | PSNR | Helped |
| Use homoscedastic Tobit regression instead of heteroscedastic Tobit regression (ablation) | Linear intensity scaling estimation · Simulated image pairs | RMSE of intensity-scaling estimates | Hurt |
| Use only intensity scaling, without alignment or low-frequency correction (ablation) | Reference-image ground-truth construction · Real image pairs | Auto-correlation of residual noise: 0.1024 -> 0.0173 | Hurt |
Closest papers by meaning
- CycleISP: Real Image Restoration via Improved Data Synthesis (CVPR 2020 · similarity 0.89)
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.89)
- NoiseDiffusion: Correcting Noise for Image Interpolation with Diffusion Models beyond Spherical Linear Interpolation (ICLR 2024 · similarity 0.87)
- Realistic Noise Synthesis with Diffusion Models (AAAI 2025 · similarity 0.87)
- Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising (CVPR 2025 · similarity 0.87)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.