基于集成GoogleNet的图像隐写分析方法  

Image steganalysis method based on integrated GoogleNet

作　　者：彭芙蓉 曹继华[1] 石和平[1] 彭斌 PENG Furong;CAO Jihua;SHI Heping;PENG Bin(School of Electronic Engineering,Tianjin University of Technology and Education,Tianjin 300222,China;College of Information Science and Technology,Beijing University of Chemical Technology,Beijing 100029,China)

机构地区：[1]天津职业技术师范大学电子工程学院,天津300222 [2]北京化工大学信息科学与技术学院,北京100029

出　　处：《天津职业技术师范大学学报》2021年第3期30-36,共7页Journal of Tianjin University of Technology and Education

基　　金：天津市自然科学基金青年项目(18JCQNJC01500).

摘　　要：针对现有的图像隐写分析方法易出现高偏差的问题,构建了一种集成GoogleNet(EGN)模型,用于图像隐写分析。该模型以GoogleNet为基础模型,采用变异的方法生成多个变异体网络,并利用集成学习思想将这些神经网络模型集成在一起,进行图像隐写分析。研究表明,该模型与现有的图像隐写分析方法和单一的GoogleNet相比,有效提高了图像隐写分析检测的准确率,对嵌入率为0.4 bpp的S-UNIWARD嵌入算法的图像隐写分析检测准确率达到96.18%。Existing image steganography analysis methods are prone to high bias.In order to fix the problem,an inte-grated GoogleNet(EGN)model is established to perform image steganalysis.The model uses GoogleNet as the base model and employs a variant approach to generate multiple variant networks,and uses integrated learning ideas to integrate these neural network models together for image steganalysis.The experimental results show that compared with the existing image steganalysis method and the single GoogleNet,the proposed model effectively improves the accuracy of image steganalysis and detection,and the accuracy of image steganalysis and detection of S-UNIWARD embedding algorithm with embedding ratio of 0.4 bpp reaches 96.18%.

关 键 词：信息隐藏 图像隐写分析 神经网络 集成学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]