基于改变率自适应分类的多类隐写分析

MULTI-CLASS STEGANALYSIS BASED ON CHANGE RATE SELF-ADAPTIVE CLASSIFICATION

一般的多类隐写分析需将每种隐写算法的各种嵌入率当作一类进行训练,因其在构造分类器时未能充分考虑算法和嵌入率对分析能力的影响,故而准确率存在一定的提升空间。提出一种基于改变率自适应分类的多类隐写分析方法,将隐写改变率和算法差异性两方面因素分层考虑。该方法使用支持向量回归法估计待测图像的改变率,进而根据改变率自适应地选择分类器,从而提高分类准确率。实验结果表明,所提方法相较于现有准确率最高的方法准确率平均提高约2%～3%,特别在嵌入率较低的情况下,提高幅度可达5%以上。

In general multi-class steganalysis, different embedding rates in each steganographic algorithm are treated as a single class for training. It does not fully take into account the impact of embedding rates and steganographic algorithms on analysis capability when constructing the classifier, so the accuracy can be improved. We propose a new approach for multi-class steganalysis based on change rate self-adaptive classification, which considers the change rates and difference of steganographic algorithms hierarchically. We use support vector regression to estimate the change rate of the testing image and then select classifiers self-adaptively according to its change rate, so that the accuracy of classification is improved. Experimental results show that this approach improves the accuracy average about 2% - 3% in comparison with current methods with highest accuracy, in particular, when the embedding rate is low, the improvement range can achieve 5% or higher.