摘要
在Haar-like特征的基础上增加新的检测特征,给出特征计算方法和积分方法,实现多角度人脸检测。将多角度人脸分为3类,即全侧脸、半侧脸和正面人脸。利用连续Adaboost算法训练各类人脸检测器,用金字塔式结构将各类人脸检测器级联成一个多角度人脸检测器。在CMU人脸检测集合上,该检测器的成功率为85.2%,高于Adaboost算法和浮点Adaboost算法。
This paper adds some new detection features on the basis of Haar-like feature, gives the feature calculation method and integration method, and achieves multi-angle face detection. It divides multi-angle face into three categories： all side face, half side face and positive face. Continuous Adaboost algorithm is used to train various types of face detector. It cascades various types of face detectors into a multi-angle face detector by using pyramid-style structure. In CMU face detection aggregation, the success rate of this detector is 85.2% which is higher than that of Adaboost algorithm and float Adahoost algorithm.
出处
《计算机工程》
CAS
CSCD
北大核心
2009年第19期195-197,共3页
Computer Engineering
基金
国家自然科学基金资助项目(10702067)
