隐写分析算法的量化评估方案  

Scenario of quantitative evaluation for steganalytic algorithms

作　　者：邓诗智[1] 刘九芬[1] 张卫明[1] 

机构地区：[1]解放军信息工程大学信息工程学院,郑州450002

出　　处：《东南大学学报（自然科学版）》2007年第A01期76-80,共5页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(60473022);河南省自然科学基金资助项目(0511011300)

摘　　要：根据隐写分析算法的特点和应用需求,结合ROC分析,提出了一个隐写分析算法的量化评估方案.包含4个评估指标,涉及算法的可靠性和准确性、适用性、分类代价和计算复杂度.并研究算法达到有效检测时,对应载密对象的最小嵌入率,分析了影响计算复杂度的因素,提出了样本量分析,并针对实际应用对虚警率和漏报率的不同需求,提出了分类代价分析.量化评估结果对于改进和应用算法具有参考意义,根据结果可以对算法的优劣进行多方面的比较,从而确定出最优的算法.According to characteristics of steganalytic algorithms as well as the need of application, a scenario of quantitative evaluation is put forward combining receiver operating characteristic （ ROC ） analysis. The scenario contains four evaluation rules relating to reliability and veracity, applicability, classifying cost and computing complexity of steganalytic algorithms. Minimum embedded ratios which algorithms can detect efficiently and the factor which influences computing complexity of algorithms are discussed. Accordingly, analysis of sample size is advanced. To adapt different requirement in false alarm rate （FAR） and false detection rate （ FDR）, analysis of classifying cost is also put forward. Evaluation results are good reference to apply and improve steganalytic algorithms. Furthermore, the results can be used for comparing algorithms different sides so that the best algorithm can be determined.

关 键 词：算法评价 隐写分析 ROC分析 LSB替换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]