Luming Liang, Sen Deng, Lionel Gueguen, Mingqiang Wei, Xinming Wu, Jing Qin
We propose a deep fully convolutional neural network with a new type of layer, named median layer, to restore images contaminated by the salt-and-pepper (s&p) noise. A median layer simply performs median filtering on all feature channels. By adding this kind of layer into some widely used fully convolutional deep neural networks, we develop an end-to-end network that removes the extremely high-level s&p noise without performing any non-trivial preprocessing tasks, which is different from all the existing literature in s&p noise removal. Experiments show that inserting median layers into a simple fully-convolutional network with the L2 loss significantly boosts the signal-to-noise ratio. Quantitative comparisons testify that our network outperforms the state-of-the-art methods with a limited amount of training data. The source code has been released for public evaluation and use (https://github.com/llmpass/medianDenoise).
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Denoising | BSD300 Noise Level 30% | PSNR | 40.9 | CNN (Median Layers) |
| Denoising | BSD300 Noise Level 50% | PSNR | 37.28 | CNN (Median Layers) |
| Denoising | Kodak24 Noise Level 50% | PSNR | 34.35 | CNN (Median Layers) |
| Denoising | Kodak24 Noise Level 70% | PSNR | 31.56 | CNN (Median Layers) |
| Denoising | BSD300 Noise Level 70% | PSNR | 32.4 | CNN (Median Layers) |
| Denoising | Kodak24 Noise Level 30% | PSNR | 36.39 | CNN (Median Layers) |
| 3D Architecture | BSD300 Noise Level 30% | PSNR | 40.9 | CNN (Median Layers) |
| 3D Architecture | BSD300 Noise Level 50% | PSNR | 37.28 | CNN (Median Layers) |
| 3D Architecture | Kodak24 Noise Level 50% | PSNR | 34.35 | CNN (Median Layers) |
| 3D Architecture | Kodak24 Noise Level 70% | PSNR | 31.56 | CNN (Median Layers) |
| 3D Architecture | BSD300 Noise Level 70% | PSNR | 32.4 | CNN (Median Layers) |
| 3D Architecture | Kodak24 Noise Level 30% | PSNR | 36.39 | CNN (Median Layers) |