基于对称α稳态模型的JPEG通用隐写分析

针对JPEG隐写术检测问题,提出一种基于DCT系数分布模型的通用隐写分析方法。根据DCT系数统计分布规律,建立双参数对称α稳态模型,并将模型参数估计问题转化为离散的函数参数优化问题,利用改进的遗传算法进行求解。提取分布模型的参数作为隐写分析的特征,并运用图像校准技术进行校准以提高特征的敏感性。设计与特征相匹配的、具有线性阶计算复杂度的单类分类器进行隐写判别。实验结果表明,该方法可以有效地检测JPEG隐写术,当嵌入率为25%时,平均检测率达到76.1%,相比传统基于模型的隐写分析方法提高5.5%,具有更高的检测性能。

JPEG Universal Steganalysis Based on Symmetric α-stable Model

Aiming at the problem of JPEG steganography detection, a universal steganalysis approach is proposed based on DCT coefficient distribution model. A better bi-parameter Symmetric α-stable（SαS） model is built according to the statistical regular of the DCT coefficient distribution. The model parameters, estimated by solving a discrete function parameter optimization problem using advanced genetic algorithm, are employed to be the steganalytic features and calibrated to improve its sensibility. A feature-matched one-class classifier with linear computation complexity is designed. Experimental results show that the method is reliable in steganalysis for all kinds of JPEG steganography. When the embedding rate is 25%, its mean detection rate achieves 76.1%, which is at least 5.5% higher than the traditional model-based steganalysis methods, it has high detection performance.