Combine Traditional Compression Method With Convolutional Neural Networks

Jianhua Hu, Ming Li, Changsheng Xia, Yundong Zhang; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 2563-2566

Abstract

Deep learning, e.g., convolutional neural networks (CNNs), has achieved great success in image processing and computer vision tasks like classification, detection and image compression. We propose a method by combining convolution neural networks and traditional compression method. The prepositive compression comes from the SVAC2(which is drafted and maintained by VimicroAI and China's Ministry of Public Security) video codec. We further improve the SVAC2 by adopting a recovering CNN network after the reconstruction. Our approach outperforms JPEG/JPEG2000/WebP standards, and is equivalent to BPG.

Related Material

[pdf]
[bibtex]
@InProceedings{Hu_2018_CVPR_Workshops,
author = {Hu, Jianhua and Li, Ming and Xia, Changsheng and Zhang, Yundong},
title = {Combine Traditional Compression Method With Convolutional Neural Networks},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}