极限学习机在图像隐写分析中的应用

为了有效提高图像隐写分析的检测正确率和速度,特结合单隐层前馈神经网络(SLFN)的特点,提出了一种基于极限学习机(ELM)的隐写分析方法.该方法首先根据Fridrich提出的多域特征提取算法从图像DCT域和空域中提取特征;得到193维原始特征;然后使用"主成份分析"法将其约简至18维;最后采用极限学习机作为分类方法构造隐写分析算法.实验表明,与目前隐写分析算法中广泛使用的支持向量机(SVM)相比,极限学习机参数调节少,学习速度快,以较少的隐层节点数取得了与SVM相似的检测正确率,能够实现针对各类JPEG图像隐写算法的有效检测.

An image steganalysis using ELM

In order to obtain higher detection rate and faster training speed for image steganalysis, a new steganalysis algorithm based on extreme learning machine（ELM） was presented by combining with the single hidden layer feedforward neural networks （SLFN）. Firstly, some features in discrete cosine transform （I）CT） and spatial domain were extracted from a JPEG image according to Fridrich＇s algorithm. Then the original 193- dimensional features were reduced to 18-dimensional features with PCA. Finally, A blindly steganalysis algorithm was constructed with the classifying technique of ELM. The experimental results showed that ELM had faster learning speed and similar classification accuracy compared with SVM since it had a smaller number of turning parameters and less number of neurons. ELM can therefore be used in a blind steganalysis for all kinds of JPEG images.