Convolutional Neural Network with Median Layers for Denoising Salt-and-Pepper Contaminations

This is a Keras version of Median Layer based CNN denoiser implemented by Luming Liang for removing salt-and-pepper noise

Prerequisites

Python 3.6 Cuda, Cudnn, ...

All python package requirements could be found in requirements.txt. Users can simply run ** pip install -r requirements.txt ** to install all required python packages.

Publication

@article{liang2021convolutional,
  title={Convolutional neural network with median layers for denoising salt-and-pepper contaminations},
  author={Liang, Luming and Deng, Seng and Gueguen, Lionel and Wei, Mingqiang and Wu, Xinming and Qin, Jing},
  journal={Neurocomputing},
  volume={442},
  pages={26--35},
  year={2021},
  publisher={Elsevier}
}

Architecture

Dataset

The training and validation datasets can be found in data directory, we use 91image as the training set, all others as validation sets

Pretrained models

The pretrained models can be found in pretrained directory

Some inference results