Research map / Image generation / Image restoration
Hyperspectral Image Denoising With Realistic Data
Tao Zhang, Ying Fu, Cheng Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| train on data from the calibrated noise model instead of complex noise model of prior work (homoscedastic Gaussian) (ablation) | 3D U-Net HSI denoiser · Real HSI test set | PSNR: 28.337 -> 29.976 | Improved |
| add signal-dependent shot noise (heteroscedastic instead of homoscedastic Gaussian) (ablation) | 3D U-Net HSI denoiser · Real HSI test set | PSNR: 29.976 -> 30.284 | Improved |
| add signal-dependent shot noise (heteroscedastic instead of homoscedastic Gaussian) (ablation) | 3D U-Net HSI denoiser · Real HSI test set | SAM: 5.8650 -> 4.4488 | Improved |
| add horizontal and vertical stripe pattern noise to the heteroscedastic model (ablation) | 3D U-Net HSI denoiser · Real HSI test set | PSNR: 30.284 -> 30.451 | Improved |
| add stripe pattern noise to the heteroscedastic model (ablation) | 3D U-Net HSI denoiser · Real HSI test set | SSIM: 0.8946 -> 0.9004 | Improved |
| add stripe pattern noise to the heteroscedastic model (ablation) | 3D U-Net HSI denoiser · Real HSI test set | SAM: 4.4488 -> 4.1364 | Improved |
| use full calibrated noise model instead of complex noise model (ablation) | 3D U-Net HSI denoiser · Real HSI test set | ERGAS: 28.831 -> 21.324 | Improved |
| train on synthetic data from our noise model instead of paired real data | 3D U-Net HSI denoiser · Real HSI test set | PSNR: 30.583 -> 30.451 | No clear effect |
| train on synthetic data from our noise model instead of paired real data | 3D U-Net HSI denoiser · Real HSI test set | SSIM: 0.8843 -> 0.9004 | Improved |
| train on synthetic data from our noise model instead of paired real data | 3D U-Net HSI denoiser · Real HSI test set | ERGAS: 20.816 -> 21.324 | Hurt |
| train the network on our realistic synthetic data instead of QRNN3D with complex noise | HSI denoiser · Real HSI test set | PSNR: 23.832 -> 30.451 | Improved |
Closest papers by meaning
- Real Noise Decoupling for Hyperspectral Image Denoising (AAAI 2026 · similarity 0.95)
- Spectral Enhanced Rectangle Transformer for Hyperspectral Image Denoising (CVPR 2023 · similarity 0.90)
- HIR-Diff: Unsupervised Hyperspectral Image Restoration Via Improved Diffusion Models (CVPR 2024 · similarity 0.90)
- Non-Local Meets Global: An Integrated Paradigm for Hyperspectral Denoising (CVPR 2019 · similarity 0.88)
- Unmixing Diffusion for Self-Supervised Hyperspectral Image Denoising (CVPR 2024 · similarity 0.88)
- Automatic Spectral Calibration of Hyperspectral Images: Method, Dataset and Benchmark (CVPR 2025 · 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.
