基于CSR-ICA模型的隐写信息盲提取算法  

Blind extraction algorithm of embedded messages based on CSR-ICA

作　　者：冯帆[1,2] 王建华[1] 张政保[2] 王惠萍[1] 戚红军[1] 张海龙[1] 

机构地区：[1]白求恩军医学院基础部,石家庄050081 [2]军械工程学院计算机工程系,石家庄050003

出　　处：《计算机工程与应用》2013年第24期70-73,共4页Computer Engineering and Applications

基　　金：河北省基金项目(No.052135;No.11213527D)

摘　　要：针对加性隐写模型,提出一种基于CSR-ICA的隐写信息盲提取算法。算法仅需一幅隐写图像,在满足ICA模型线性约束条件下得到载体信号的估计信号,通过Contourlet稀疏性表示(CSR)对模型输入信号进行前置处理,优化选取归一化峭度性较大的信号作为模型输入信号,将归一化峭度作为分离算法学习的目标函数,避免异常值给分离算法带来的误差。算法具有较好的综合性能,并且克服了Chandramouli算法的局限性,提取正确率平均为90%。仿真实验结果给出了算法的有效性验证。Aiming at steganography adding model, a blind extraction algorithm of embedded messages is proposed based on CSR-ICA. It attains estimated signal of original cover signals using linear combination just one digital image, which consists with constrain of ICA model. Then signals with optimal normalization kurtosis are selected and inputted to ICA model by optimization algorithm, In order to eliminate effect of outlier, normalization kurtosis is taken as objective function of separation algorithm on precision. The algorithm has higher comprehensive performance and conquers the localization of proposed algorithm by Chan- dramouli. Average extraction correct rate is 90%. The results of simulation experiments prove its validity further.

关 键 词：Contourlet稀疏性表示(CSR) 独立成分分析(ICA)模型 隐写 盲提取 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]