基于多方向混合滤波器的轻量化图像隐写分析模型  

A Lightweight Image Steganalysis Model Based on Multi-directional Hybrid Filters

作　　者：胡原平 阎红灿 刘盈 Hu Yuanping;Yan Hongcan;Liu Ying(School of Science,North China University of Science and Technology,Tangshan,Hebei 063210;Hebei Key Laboratory of Data Science and Application(North China University of Science and Technology),Tangshan,Hebei 063210)

机构地区：[1]华北理工大学理学院,河北唐山063210 [2]河北省数据科学与应用重点实验室(华北理工大学),河北唐山063210

出　　处：《信息安全研究》2025年第4期318-325,共8页Journal of Information Security Research

基　　金：教育部协同育人项目(202102269033)。

摘　　要：针对当前图像隐写分析模型参数量庞大、泛化能力有限、准确率不高等问题,构建了一个基于多方向混合滤波器的轻量化图像隐写分析模型.该模型设计多方向、多尺寸的高低频混合滤波器组并应用通道注意力机制对图像进行预处理,自适应地提取图像中有效特征,提高模型的泛化能力;特征提取模块设计包含残差模块的多层卷积,对图像特征进行深度提取,提高模型对特征的捕捉能力;降维模块采用深度可分离卷积代替传统卷积,有效降低模型参数量,实现轻量化.实验数据对比分析表明,该模型具有轻量化特点和较好的泛化能力,同时提高了隐写分析的准确性.Aiming at the problems of large number of parameters,limited generalization ability and low accuracy of current image steganalysis model,a lightweight image steganalysis model based on multi-direction hybrid filters is constructed.In this model,a multi-directional and multi-size high and low frequency hybrid filter bank is designed and the channel attention mechanism is used to preprocess the image,so as to adaptively extract the effective features in the image and improve the generalization ability of the model.The feature extraction module designs a multi-layer convolution including the residual module to extract the image features in depth and improve the ability of the model to capture features.The dimensionality reduction module adopts depthwise separable convolution instead of traditional convolution,which effectively reduces the number of model parameters and achieves lightweight.Comparative analysis of experimental data showed that the model had the characteristics of lightweight and good generalization ability,and improved the accuracy of steganalysis.

关 键 词：图像隐写分析 混合滤波器 通道注意力机制 深层卷积 深度可分离卷积 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]