SIDUMN: Self-Supervised Image Denoising Under Symmetric Multiplicative Noise
Organizations: Department of Mathematics, Kyungpook National University, Korea · Department of Mathematics, Yeungnam University, Korea
Abstract
Self-supervised denoising is challenging when the corruption is multiplicative and signal-dependent. In this letter, we propose SIDUMN, a symmetric noisier framework for self-supervised denoising under mean-one symmetric multiplicative noise. Given noisy observations and the noise law, the method samples an independent synthetic noise factor from the same law and constructs two complementary noisier images by applying the factor and its reflection about one. The denoising network is trained to predict one noisier image from the other, and at inference the reconstructions of such antithetic pairs are averaged. We show that the population target of the training loss is the clean conditional mean minus a multiplicative cross-term. We further prove that, for any trained network, antithetic averaging never increases the expected squared error of a single noisier-input reconstruction. Numerical results validate the theoretical findings and the proposed method.
Figures & tables
| BSD300 | |||
| Noisy input | 23.40 / 0.6269 | 21.00 / 0.5440 | 19.17 / 0.4763 |
| DIP [ 25 ] | 28.49 / 0.8179 | 27.68 / 0.7924 | 26.96 / 0.7683 |
| Noise2Self [ 8 ] | 30.00 / 0.8977 | 29.69 / 0.8783 | 29.06 / 0.8597 |
| LogNN2N [ 12 , 16 ] | 30.92 / 0.8860 | 29.16 / 0.8469 | 27.83 / 0.8153 |
| SIDUMN | 32.23 / 0.9164 | 30.71 / 0.8868 | 29.23 / 0.8412 |
| SIDUMN (Oracle) | 32.29 / 0.9187 | 30.78 / 0.8899 | 29.43 / 0.8520 |
| BSD300 | |||
| 31.46 / 0.8921 | 29.60 / 0.8437 | 27.74 / 0.7652 | |
| 30.76 / 0.8929 | 29.28 / 0.8534 | 27.74 / 0.7858 | |
| Avg. | 32.17 / 0.9148 | 30.61 / 0.8825 | 29.00 / 0.8275 |
| Avg. | 32.20 / 0.9156 | 30.66 / 0.8847 | 29.12 / 0.8343 |
| Avg. | 32.21 / 0.9160 | 30.69 / 0.8858 | 29.17 / 0.8379 |
| Avg. | 32.22 / 0.9163 | 30.70 / 0.8864 | 29.20 / 0.8398 |
Self-Supervised Image Denoising
Self-supervised image denoising learns to remove noise from images without requiring clean training targets.
See all 28 papers