基于深度学习的数字图像隐写和隐写分析术研究概述

Study Overview of Digital Image Steganography and Steganalysis Based on Deep Learning

隐写术的不断发展使得隐写分析术面临的挑战越来越大。近年来,有不少学者围绕深度学习网络强大的图像特征表达学习能力进行隐写分析术研究,从而避开人工设计分类特征,减少人的参与度,用数据和算法驱动机器去实现数字图像是否含密的判定。本文将从数字图像的全局和局部统计分布特性这两个方面出发,梳理传统的和基于深度学习的隐写与隐写分析术在空域和JPEG域上的研究进展,并对数字图像隐写和隐写分析术未来发展方向做简要讨论。

With the continuous development of steganography,steganalysis is facing more and more challenges.In recent years,many scholars have carried out steganalysis research around the powerful image feature expression learning ability of deep learning network,so as to avoid manually designing classification features,reduce people’s participation,and drive the machine with data and algorithm to determine whether the digital image contains secret.Starting from the global and local statistical distribution characteristics of digital images,this paper combs the research progress of traditional and deep learning based steganography and steganalysisg in spatial domain and JPEG domain,and briefly discusses the future development direction of digital image steganography and steganalysis.