CAJ | 학술논문

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。
침대 JPEG 은사술검측문제，제출일충기우 DCT 계수분포모형적통용은사분석방법。근거 DCT 계수통계분포규률，건립쌍삼수대칭α은태모형，병장모형삼수고계문제전화위리산적함수삼수우화문제，이용개진적유전산법진행구해。제취분포모형적삼수작위은사분석적특정，병운용도상교준기술진행교준이제고특정적민감성。설계여특정상필배적、구유선성계계산복잡도적단류분류기진행은사판별。실험결과표명，해방법가이유효지검측 JPEG 은사술，당감입솔위25%시，평균검측솔체도76.1%，상비전통기우모형적은사분석방법제고5.5%，구유경고적검측성능。
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.