The new situation that in recent years the sharp increase of steganalysis features dimension causes the problems such as the curse of dimension of blind quantitative steganalysers.In order to solve the problems,we propose a blind quantitative steganalysis method based on ensemble regressions.The method chooses linear least square regression (LLSR) to reduce computational complexity,and uses bootstrap re-sampling method through ensemble classifier technique to generate different training sets for enhancing the diversity of regressions and guaranteeing the accuracy rate,then it randomly chooses the part of original features to train and predict in order to shorten the runtime.Experimental results show that compared with current best methods,our method decreases the prediction error by about 70％ of theirs and shortens the runtime by about 5％ of theirs.%针对近年来隐写分析特征维度激增的新情况,及其带来的定量隐写分析预测器维度之灾等问题,提出一种基于组合回归的盲定量隐写分析方法.该方法选择线性最小二乘回归LLSR(Linear Least Square Regression)降低计算复杂度,并通过组合分类器技术在多次迭代中使用自助法(bootstrap)重采样技术生成不同的训练集,以提高回归的多样性进而保证准确率,然后随机选择特征的一部分做训练和预测,以缩短执行时间.实验结果表明,该方法与现有最优方法相比,预测误差降低至现有方法的70％,执行时间缩短至现有方法的5％左右.
展开▼
