摘要：

针对高维大样本空间中支持向量机(SVM)存在计算复杂度高、分类精度低等问题,在随机子空间方法与主成分分析方法的基础上,提出一种特征加权支持向量机的高维隐写盲检测方法。通过随机子空间对原始高维样本的特征空间进行随机采样,产生多个低维的特征子集,在特征子集中采用主成分分析法进行特征提取,并利用卡方统计计算特征权重,运用特征加权核函数训练各基SVM 分类器,并用多数投票法融合各基分类器结果得到最终分类结果。对HUGO 隐写算法的实验结果表明,该方法能有效降低SVM 计算复杂度,与传统方法相比,具有较高的隐写检测率和更快的分类速度。

关键词: 隐写检测, 随机子空间方法, 主成分分析, 支持向量机, 高维特征

Abstract:

Aiming at the problem for high-dimensional and large-sized data,such as high computational complexity and low classification accuracy,this paper proposes a high dimensional steganography blind detection method based on Random Subspace Method and Principal Component Analysis (RSM-PCA) feature weighted Support Vector Machine (SVM). The method selects features randomly from original high-dimensional features to form feature subsets using random subspace method,the principal component analysis is adopted to carry out feature extraction on feature subsets and the chi-square statistic is adopted to measure the weights of extracted features,the feature weighted kernel function is used to train SVM and the finial decision is yielded using the majority vote method. Experimental results about HUGO steganographic algorithm show that this method can effectively reduce the computational complexity of SVM,compared with traditional algorithms,this method effectively improves the detection rate of steganalysis in JPEG images and achieves faster speed of image classification.

Key words: stego-detection, Random Subspace Method(RSM), Principal Component Analysis(PCA), Support Vector Machine(SVM);high-dimensional feature

中图分类号: 

• TP309