摘要： Adaboost算法采用单阈值弱分类器，难以拟合复杂分布，其训练过程收敛速度较慢。针对该问题设计一种多阈值弱学习器，利用平方和减少最大化准则划分节点并生成弱分类器，在训练数据集上采用GAB算法将弱分类器提升为强分类器。实验结果表明，在弱分类器数目相同的情况下，该方法的正样本误报率低于Adaboost算法。

关键词: 人脸检测, boosting方法, 实值Adaboost, 平缓Adaboost

Abstract: Aiming at the problem that the Adaboost algorithm using simple threshold weak classifiers is too weak to fit complex distributions and its slow convergence rate in training, this paper designs a multiple thresholds weak learner. This learner splits the nodes by using biggest reduction in the sum of squares as the partition criteria and builds a weak classifier. It boosts weak classifiers using GAB algorithm on training dataset. Experimental results show that the false positive rate of this method is lower than Adaboost algorithm at the same number of weak classifiers.

Key words: face detection, Boosting method, real Adaboost, gentle Adaboost

中图分类号: 

• TP391