ImageNet Pre-trained CNNs for JPEG Steganalysis

ImageNet Pre-trained CNNs for JPEG Steganalysis
复制标题

DOI:
10.1109/wifs49906.2020.9360897
复制
发表时间:
2020-12
期刊:
2020 IEEE International Workshop on Information Forensics and Security (WIFS)
影响因子:
--
通讯作者:
Yassine Yousfi;Jan Butora;Eugene Khvedchenya;J. Fridrich
Yassine Yousfi;Jan Butora;Eugene Khvedchenya;J. Fridrich
中科院分区:
其他
文献类型:
--
作者:
Yassine Yousfi;Jan Butora;Eugene Khvedchenya;J. Fridrich

文献摘要

被引文献

相似文献

在本文中,我们研究了预训练的计算机视觉深度架构,例如用于隐写分析的 EfficientNet、MixNet 和 ResNet。这些在 ImageNet 上预训练的模型可以相当快速地针对 JPEG 隐写分析进行细化,同时提供比专门用于隐写分析的 CNN(例如从头开始训练的 SRNet)更好的性能。我们展示了不同架构如何在 ALASKA II 数据集上进行比较。我们证明,正如其他顶级竞争对手所注意到的那样,避免第一层中的池化/跨步可以实现更好的性能,这与许多专为隐写分析而设计的 CNN 的设计选择是一致的。我们还展示了预训练的计算机视觉深度架构如何在 ALASKA I 数据集上执行。
In this paper, we investigate pre-trained computer-vision deep architectures, such as the EfficientNet, MixNet, and ResNet for steganalysis. These models pre-trained on ImageNet can be rather quickly refined for JPEG steganalysis while offering significantly better performance than CNNs designed purposely for steganalysis, such as the SRNet, trained from scratch. We show how different architectures compare on the ALASKA II dataset. We demonstrate that avoiding pooling/stride in the first layers enables better performance, as noticed by other top competitors, which aligns with the design choices of many CNNs designed for steganalysis. We also show how pre-trained computer-vision deep architectures perform on the ALASKA I dataset.