【摘要】 近年来,图像隐写分析技术与机器学习的结合越来越紧密.字典学习作为机器学习中的一个重要研究方向,在解决众多实际问题时展现了其独特的优势,但在隐写分析领域中的研究成果却十分有限.针对隐写方法失配和嵌入率失配2个隐写分析领域亟待解决的问题,提出了基于混合字典学习的隐写分析算法.该算法为实现对载体图与隐密图之间的差异进行编码,分别为2类图像学习各自的子字典,同时为表示载体图与隐密图之间的相同之处,单独学习一个共享字典.此外,该算法同时学习出综合型字典和解析型字典,分别以重构数据和编码分类.实验结果表明,所提出的基于混合字典学习的隐写分析方法具有很好的性能.更多还原

【Abstract】 Steganalysis technology is more and more close to machine learning in recent years.Dictionary learning,as a crucial research domain in machine learning,shows the unique advantages in solving numerous practical problems,but its research results in steganalysis domain are limited.To address the mismatch of steganography and embedding rate which are needed to be solved in steganalysis domain,we propose a steganalysis algorithm based on hybrid dictionary learning.In order to encode the differences between cover images and stego images,our algorithm learns specific sub-dictionaries for two kinds of images,at the same time,a shared dictionary is learned for representing the common content.Except it,we learn a synthesis dictionary for data reconstruction,and an analysis dictionary for classification.The experimental results demonstrate the steganalysis based on hybrid dictionary learning has high performance.更多还原