中文摘要：

为了解决多视角人脸检测中多视角导致的人脸结构不同的问题和人脸与非人脸之间的误分类风险不同的问题,检测特征使用局部二值模式（LBP）及统计直方图,人脸非人脸分类器使用可控风险敏感AdaBoost（CCS-Ada-Boost）.LBP及统计直方图能够描述多视角的人脸结构;CCS-AdaBoost能够在降低总体的误分类风险的同时最小化分类错误率.实验中,LBP特征的性能在正面人脸检测上比Haar-like特征更好.CCS-AdaBoost分类器在一定条件下也比普通AdaBoost分类器有更好的性能,并且弥补了风险敏感AdaBoost分类器（CS-AdaBoost）对靠近分类边界的样本分类不好的缺陷.最终的多视角人脸检测器在CMU-Profile测试集上获得了满意的结果.该算法实现了鲁棒的多视角人脸检测方法,在相同虚警率下获得比其他人脸检测方法更好的结果,能够有效地解决多视角人脸检测中的2个问题.

英文摘要：

Detection features used local binary pattern（LBP） and its histogram in order to solve the problems in multi-view face detection of different face structure caused by multi-views and different misclassification cost between faces and non-faces.The face and non-face classifiers used controlled cost-sensitive AdaBoost（CCS-AdaBoost）.LBP and its histogram can describe the multiview face structure;CCS-AdaBoost can minimize the total misclassification cost while minimizing the classification error rate.In the experiments,the LBP feature＇s performance in frontal face detection is better than Haar-like features.CCS-AdaBoost also had a better performance than the original AdaBoost algorithm under certain conditions and overcame the cost-sensitive AdaBoost（CS-AdaBoost） algorithm＇s drawbacks of classifying the samples near the classification boundary not well.The final multiview face detector had a satisfied detection result in CMU-Profile test set.The proposed method realized a robust multiview faces detection algorithm,and got better results than other face detection methods at the same false alarm rate.The method can effectively solve the foregoing two problems in multiview face detection.