论文类型：期刊论文
发表刊物：CHINESE JOURNAL OF ELECTRONICS
收录刊物：Scopus、EI、SCIE
卷号：18
期号：3
页面范围：455-459
ISSN号：1022-4653
关键字：Steganography; Universal; Quantitative setanalysis; Statistical learning; Support vector regression (SVR)
摘要：The objective of quantitative steganalysis is to achieve reliable estimation of embedded message length of a suspected digital object. This type of methods has received considerable attention due to its capability of providing more detailed information about embedded secrets rather than just determining whether a suspected object is stego or not. In this paper, we present the methodology of "universal quantitative steganalysis", which is a practical, unified approach for designing quantitative steganalytic methods based on statistical learning techniques. This methodology models the relation between embedded message length and statistical feature change caused by the embedding process with a multivariable function, and solves the problem of optimal parameter estimation with SVR (Support vector regression) technique. Experimental results indicate that new quantitative steganalytic methods applying the presented methodology can achieve excellent performance for F5 and MB1 steganographic mechanisms.