公开(公告)号：US09582710B2

公开(公告)日：2017-02-28

申请号：US14558414

申请日：2014-12-02

Applicant: Huawei Technologies Co., Ltd.

Inventor： Yu Zhang ,  Zhijun Du ,  Minjie Wang

IPC: G06K9/00 ,  G06K9/62 ,  G06K9/46

CPC classification number: G06K9/00295 ,  G06K9/00221 ,  G06K9/4614 ,  G06K9/6256 ,  G06K9/6257 ,  G06K2009/4666

Abstract: Embodiments of the present invention disclose methods for generating a strong classifier for face detection. The methods include determining, according to a size of a prestored image training sample, a parameter of weak classifier of the image training sample, obtaining a sketch value of each of the weak classifiers of the image training sample, calculating a weighted classification error of each of the weak classifiers according to the sketch value and an initial weight of the image training sample, obtaining at least one optimal weak classifier according to the weighted classification error, and generating a strong classifier for face detection according to the optimal weak classifiers. The embodiments of the present invention further disclose an apparatus for generating a strong classifier for face detection. The embodiments of the present invention have advantages of improving robustness of code against noise and reducing a false detection rate of face detection.

Abstract translation: 本发明的实施例公开了用于生成用于面部检测的强分类器的方法。 所述方法包括根据预先存储的图像训练样本的大小来确定图像训练样本的弱分类器的参数，获得图像训练样本的每个弱分类器的草图值，计算每个图像训练样本的加权分类误差 根据草图值和图像训练样本的初始权重，根据加权分类误差获得至少一个最优弱分类器，并根据最优弱分类器生成用于面部检测的强分类器。 本发明的实施例还公开了一种用于生成用于面部检测的强分类器的装置。 本发明的实施例具有改善代码对噪声的鲁棒性并降低面部检测的错误检测率的优点。