Abstract: Face detection method based on the cascade architecture using harr-like feature has high detection rate and rapid detection speed.But harr-like feature is sensitive for edge and lines and can describe given direction.Combining cascade decision-making technology，the face detection method based on boosting covariance features is presented.Firstly，the covariance features are figured out.And then，the weak detectors are constructed using those features.Finally，utilizing the Adaboost algorithm to combine these decision results from the weak detectors，the face images are filtered through the waterfall cascade architecture.Consequently，the final decision outcome are gained.The experimental results testify the detector has anti-noise power.The detection rate is higher than the former methods based on the cascade architecture using harr-like feature obviously.

Key words: face detection, covariance feature, Adaboost, feature extraction

摘要： 基于Harr式特征分层筛选的人脸检测方法速度快、检测率高。但Harr式特征对边缘、线段比较敏感，只能描述特定走向的图形结构。结合分层筛选技术，提出了Boosting协方差特征人脸检测方法。该方法先计算协方差矩阵特征，然后由这些特征构造弱分类器，最后借助Adaboost方法组合这些弱分类器的输出结果来对测试图片进行瀑布式分层筛选，从而获得最终判决结果。测试实验显示所提方法具有较强的抗噪能力，检测率相比原基于Harr式特征分层筛选的方法有显著提高。

关键词: 人脸检测, 协方差特征, Adaboost方法, 特征提取