基于改进AdaBoost的快速人脸检测算法

RAPID FACE DETECTION ALGORITHM BASED ON IMPROVED ADABOOST

针对传统AdaBoost用于人脸检测时需要的特征数目多,检测速度慢的问题,提出一种基于改进AdaBoost的快速人脸检测算法。一方面,提出使用双阈值的弱分类器代替传统的单阈值弱分类器,提高单个特征的分类能力;另一方面,引入信息熵作为特征相关度的度量方法,在特征选择时每一轮循环中只选择与已选出特征相关度较低的特征,从而减少特征之间的冗余信息。实验结果表明,相对于传统AdaBoost人脸检测算法,该方法使用较少的特征即可达到较高的检测准确率,检测速度得到显著提高。

When applying in face detection, traditional AdaBoost has the problems of asking many feature numbers and slow speed in detection. In light of this, a rapid face detection algorithm based on improved AdaBoost is proposed. On the one＇hand, dual-threshold weak classifiers are used to replace the traditional single-threshold weak classifier and this has improved the classification capability on single feature. On the other hand, the information entropy is introduced as the metric means of feature relevance, during the feature selection, in each round of cycle only those features with low feature relevance to the selected features will be chosen, therefore the redundant information between the features is reduced. Experimental results show that compared with traditional AdaBoost face detection algorithm, this one can achieve higher detection correct rate using less features, and the detection speed is magnificently enhanced.