摘要： 提出一种由若干个支持向量机(SVM)分类器串连而成的级联SVM算法，用于人脸图像性别识别。该算法把容易被前一层分类器分类的训练样本过滤掉，将难度较高的训练样本重新组织起来训练新一层的分类器。结合级联分类器和SVM理论的优势，在训练过程中能够使用更多的样本，具有更好的识别性能。在同一硬件实验条件下的实验结果表明，单层SVM最多只能训练7万样本，而四层级联SVM的训练样本数可达12万以上，相应的识别率也从96.6%上升至98.4%。

关键词: 统计学习, 支持向量机, 分类器, 大样本训练, 级联, 性别识别

Abstract: Support Vector Machine(SVM) is a popular statistical learning methods, but large-scale training of SVM is limited by hardware. This paper proposes a face image gender classification algorithm based on a cascade connection SVM, which filters the easily classified samples by pre-layer classifiers, and re-organizes the left tough samples to train the next SVM layer. Meanwhile, more samples are used, and the classifier has better recognition performance. Experimental results under the same hardware conditions show that only 70 000 samples can be contained one time to train one-layer SVM, while more than 120 000 samples are involved in four-layer SVM, the corresponding recognition rate is 96.6% to 98.4%.

Key words: statistical learning, Support Vector Machine(SVM), classifier, large-scale training, cascade connection, gender identification

中图分类号: 

• TP391