基于支持向量AAM迭代学习的性别分类算法

为了提高性别检测的精度，提出了一种支持向量机（SVM）与主动外观模型（AAM）相结合的迭代学习算法.采用AAM对初始训练样本建模，在此基础上构造SVM分类器.在当前迭代过程所产生的支持向量中随机选择不同性别的样本，对其AAM参数线性插值而生成一系列伪样本，并从中选取被当前分类器误分类或正确分类但分类可信度低的样本参与下次迭代学习.实验结果表明，采用该算法所构造的伪样本是初始训练样本的有效补充，提出的伪样本选择方案优于传统的Bootstrap方法，迭代学习方法逐步提高了性别分类器的检测精度.

Support vector AAM based iterative learning algorithm for gender classification

In order to increase accuracy in gender classification,an iterative learning approach combining support vector machine(SVM) and active appearance model(AAM) was proposed.The original training examples were modeled by using AAM before the SVM classifier learning process.During the current iteration,some pairs of support vectors with different gender were selected randomly and their AAM parameters were interpolated to generate a series of pseudo-examples.Only the pseudo-examples classified incorrectly by the current classifier or classified correctly but with low confidence were selected for the next iterative learning.Experimental results show that the pseudo-examples created by this algorithm complement effectively the original training examples,that the proposed pseudo-example selecting scheme outperforms the conventional Bootstrap method,and that iterative learning approach upgrades the gender detection accuracy stepwise.