基于实验的典型隐写通用盲检测方法性能分析

Experiments-Based Performance Analysis of Typical Universal Blind Steganalysis Methods

现有文献中存在多种数字图像信息隐写通用盲检测方法,其检测正确率和适用的范围各不相同。基于同一实验图像库,对现有的典型隐写通用盲检测方法的性能进行了一系列实验分析。涉及到的盲检测方法包括质心法、概率密度函数矩法、特征函数矩法、共生矩阵法、两域特征联合法、三域特征综合法和基于小波包分解的检测方法等。从实验的角度给出了小波包分解特征法性能最优、多域特征提取方法更适合与BP(Back Propagation)神经网络分类器搭配使用等结论。

Existing universal digital image blind steganalysis methods vary in detection accuracy and application scope. Based on the same image library, this paper presents experimental analysis and compares the performance of typical universal blind steganalysis methods. The typical blind detection methods include the COM （Center of Mass） method, the PDF （probability density function） moment method, the CF （characteristic function） moment method, the co-occurrence matrix method, the combined method of two-domain characteristics, the combined method of three-domain character- istics and the wavelet packet-based blind detection method. The results indicate that the wavelet packet feature method exhibits the best performance, and the multi-domain feature extraction method is more suitable to be used in concert with the BP （Back Propagation） neural network classifier.