基于集成GoogleNet的图像隐写分析方法

Image steganalysis method based on integrated GoogleNet

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%.