Abstract: Dr Viola puts forward a fast face detection algorithm based on Haar-like features，which is promising.Though the research on it，it finds the algorithm still have two aspects to improve.Firstly，it’s difficult to give out a reasonable tradeoff between the detection rate and the false positive rate.Secondly，the face posture which can be detected depends on the training samples greatly.In the paper，it puts forward improving measures according Viola’s，and designs a face detection system which can detect the faces with an angle not more than ±45° compared with the upright.

Key words: face detection, Haar-like feature, skin model, cascade classifier, strong classifier, weak classifier

摘要： 美国的Voila博士提出的基于Haar-like特征的人脸检测算法是一种具有巨大发展潜力的新算法，快速而准确。通过研究，认为这一算法还存在两个有待改进的地方：一是在检测率和误检率之间难以权衡，二是可检测人脸姿态受训练样本制约。对此提出了改进措施，设计了一个可以检测相对于垂直方向有±45°偏转的正面人脸的检测算法，与基于Haar-like特征的人脸检测算法相比，具有更好的鲁棒性和更低的误差率。

关键词: 人脸检测, Haar-like特征, 肤色模型, 级联分类器, 强分类器, 弱分类器