README.md

Enhanced CNN for image denoising by Chunwei Tian, Yong Xu, Lunke Fei, Junqian Wang, Jie Wen and Nan Luo is published in CAAI Transactions on Intelligence Technology (SCI-IF:7.985), 2019. It is implemented by Pytorch.

It is an excellent paper for CAAI Transactions on Intelligence Technology in 2020. The code of ECNDNet is collected by Profillic (The largest collection of ML models and code to power your projects) at https://www.catalyzex.com/paper/arxiv:1810.11834.

This code written with Pytorch>=0.4.

1. Dependences

pyTorch(>=0.4)

torchvision

openCv for Python

HDF5 for Python

Python 2.73

2. Test ECNDNet

If the noise level is 15, we will run the following commod:

python test.py --num_of_layers 17 --logdir sigma15/ --test_data Set68 --test_noiseL 15

or python test.py --num_of_layers 17 --logdir sigma15/ --test_data Set12 --test_noiseL 15

If the noise level is 25, we will run the following commod:

python test.py --num_of_layers 17 --logdir sigma25/ --test_data Set68 --test_noiseL 25

or python test.py --num_of_layers 17 --logdir sigma25/ --test_data Set12 --test_noiseL 25

If the noise level is 25, we will run the following commod:

python test.py --num_of_layers 17 --logdir sigma50/ --test_data Set68 --test_noiseL 50

or python test.py --num_of_layers 17 --logdir sigma50/ --test_data Set12 --test_noiseL 50

3. Network architecture

4. Experiment results

ECNDNet for BSD68

ECNDNet for Set12

Running time of ECNDNet for a noisy image of different sizes

Visual results of ECNDNet for BSD68

Visual results of ECNDNet for Set12

If you cite this paper, please refer to the following format:

1. Tian C, Xu Y, Fei L, et al. Enhanced CNN for image denoising[J]. CAAI Transactions on Intelligence Technology, 2019, 4(1): 17-23.

2. @article{tian2019enhanced,

title={Enhanced CNN for image denoising},

author={Tian, Chunwei and Xu, Yong and Fei, Lunke and Wang, Junqian and Wen, Jie and Luo, Nan},

journal={CAAI Transactions on Intelligence Technology},

volume={4},

number={1},

pages={17--23},

year={2019},

publisher={IET}

}