https://arxiv.org/pdf/1803.05787.pdf
This is a simple implementation of our work "DeepN-JPEG: A Deep Neural Network Favorable JPEG-based Image Compression Framework"
- We implement frequency analysis modul for one image (can be extended to multiple images)
- You need to map the generated matrix to quantization table and replace the original one in JPEG framework. e.g.:
./cjpeg -dct int -qtable qt.txt -baseline -opt -outfile $C temp.bmp
- If you use this code please cite our paper:
@inproceedings{liu2018deepn,
title={DeepN-JPEG: A deep neural network favorable JPEG-based image compression framework},
author={Liu, Zihao and Liu, Tao and Wen, Wujie and Jiang, Lei and Xu, Jie and Wang, Yanzhi and Quan, Gang},
booktitle={Proceedings of the 55th Annual Design Automation Conference},
pages={18},
year={2018},
organization={ACM}
}
python DeepN.py