采用L-M算法的JPEG图像隐写分析

Steganalysis for JPEG images based on L-M algorithm

提出了一种针对JPEG图像的通用隐写分析系统。首先对实验图像提取其在小波域的细节部分和近似部分系数的特征函数矩和实验图像的第一层小波分解后的对角子带D1再次进行分解所得到的细节部分和近似部分系数的特征函数矩作特征的有效性分析,通过对有效特征的选择和提取,得到一组训练集合,最后采用基于L-M算法的BP神经网络来进行分类。实验结果表明,这是一种有效的、高精度的盲检测方法,能够准确识别出JPEG图像是否含有隐密信息。

This paper proposes a universal steganalysis system for JPEG images.Several kinds of steganography algorithms are used for embedding the secret information into images.Then this paper effective analyzes the experiment of the characteristic function for the coefficients of the wavelet decomposition details and approximation components.This paper also further decomposes the first scale diagonal D1 subbands and obtains the coefficients of the details and approximation components.Then does the same effectiveness analyses experiment for the characteristic function moment of the coefficients.The L-M algorithm BP neural network is adopted to train the feature set.This paper gets better detection results through the experiment for JPEG images.