Abstract: This paper proposes a method to identify hiding domains (including the pixel, DCT, and DWT domains), constructs image features. Multi-class Support Vector Machine(SVM) is trained based on image features. The method can detect the presence of hidden message, and identify the hiding domains. Experimental results show that, the distinguish ratio rises while the embed capacity rising; the distinguish ratio can reach 85% when the embed capacity is 20%.

Key words: blind steganalysis, hiding domain, multi-class

摘要： 提出一种区分隐写域(包括像素域、DCT域、DWT域)的盲检测方法，构造图像特征向量，建立一个多分类的支持向量机，根据特征向量对图像进行训练。该方法能够识别隐藏信息和其隐写域。实验结果表明，当嵌入容量达到20%时，识别率提高到85%以上。

关键词: 盲检测, 隐写域, 多分类

CLC Number: 

• TP391