DCT系数的统计模型及其接收机性能分析  

Statistical Modeling of DCT Coefficients and Detector Performance in Terms of Receiver Operation Characteristic Curves

作　　者：毛家发[1,2] 林家骏[1] 韩津生[1] 

机构地区：[1]华东理工大学信息科学与工程学院,上海20023 [2]上饶师范学院数学计算机系,江西334001

出　　处：《中国图象图形学报》2009年第9期1782-1789,共8页Journal of Image and Graphics

基　　金：江西省教育厅科技基金项目(Gjj08462);上海市教育发展基金会晨光计划项目(2008CGB21)

摘　　要：隐写分析技术已经成为网络信息安全的一个重要方向,如何有效地监控和检测网络中传递的秘密信息对于保障国家安全具有重要现实意义。针对国内外学者提出各种不同的DCT系数分布模型,提出了图像的DCT系数α稳态簇模型。考虑数据的实际统计特征,系统地描述了图像DCT系数的α稳态簇模型,验证了图像DCT系数统计模型既不服从于广义高斯模型、高斯模型、Lapalacian模型,而是服从于α稳态簇模型,因为α稳态簇模型具有足够的柔韧性,随着特征指数α改变,其分布形状就会改变。同时,根据工程上简易、实用、可行性和性能好的要求,提出了一种新颖的、α稳态簇模型之一的、柯西接收机的隐藏信息盲检测方法。这种检测接收机与广义高斯接收机相比,其性能更具鲁棒性,适应性更广。与文献[8]相比,其性能不但没有下降,有的反而提高了,而且运行速度也快了几十倍。Steganalysis technique is of great importance to network information security and how to search and detect secret messages transferred in network is crucial and practical to safeguard national security. Against a variety of DCT coefficients distribution model proposed by the scholars at home and abroad, this paper introduces the α-stable family models of the image＇ s DCT coefficients. Take into account real statistical properties, this paper also describes the α-stable family models of cover image＇ s DCT domain by the numbers. And we also testifieed by the experiment that image ＇ s DCT coefficients satisfy α-stable family distribution but GGD（ generalized Gaussian distribution） and Gaussian distribution and Lapalacian distribution. The α-stable family models have enough flexibility, and their shape of distribution changes with characteristic index α. At same time, this paper designs a new processor to blindly hide information detection using Cauchy receiver operating machine based on project needs： facility, practicality, feasibility and performance good. This detection performance is better than GGD ＇ s detection performance. Compared to the reference [ 8 ] , our model performance enhanced. The operating speed is increased by scores of times.

关 键 词：隐写分析 α稳态簇模型 接收机 ROC曲线 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]