基于Adaboost算法的多角度人脸检测

文中提出了一种基于Adaboost算法的多角度人脸检测方法.多角度人脸检测问题的研究与正面人脸检测相比,相对薄弱,离实际应用的需求还比较远.首先使用Haar特征设计并构造弱分类器空间,用Adaboost算法学习得到基于视图的多分类器级联的人脸检测器;然后将多角度人脸划分成三类:全侧脸,半侧脸及正面人脸,并为不同角度的人脸建立不同的检测器分别用于检测.在CMU侧面人脸检测集合上,用基于Adaboost的方法对多角度人脸图像进行仿真实验,检测正确率为89.8%,误报数为243个.相比Schneiderman等人的方法,该方法具有更好的性能.

Multi-View Face Detection Based on Adaboost Algorithm

This paper presents a multi-view face detection method based on Adaboost algorithm.Compared with frontal-view face detection researches,multi-view face detection's are much weaker and further from practical application because of many difficulties.This paper designs weak classifiers using Haar-like features,and correspondingly constructs the space of weak classifiers from the cascade face detector,which is learnt by using Adaboost algorithm.They categorize multi-view faces into 3 different types:frontal faces,half-profile views face and full-profile views face,and design detector for each type respectively;then experiment their method upon multi-view faces of CMU profile face test set.Result shows an 89.8% correct with 243 errors.Compared with the result of Shneiderman's method,this approach has better performance.