基于分级式特征提取的多视角人脸检测算法

Multi-view face detection by hierarchical feature extraction

提出了一种基于分级式特征提取的多视角人脸检测算法.首先,将训练所用人脸样本按照视角进行分组;其次,分别对每组样本进行特征提取,针对单一特征的局限性,提出了梯度方向直方图（HOG）和局部二值模式（LBP）的融合特征,为了快速构建特征金字塔,提出了一种分级式特征提取的方法;再次,使用基于隐含变量的支持向量机（LSVM）训练模型参数,获得多个模型;最后,将这些模型组合起来构成混合模型.在FDDB和AFW人脸数据库上进行了实验,结果表明：本算法可实现复杂背景下的多视角人脸检测,且比现有算法效果更好.

A face detection algorithm based on hierarchical feature extraction was proposed.Firstly,the face samples in the training set were divided into several groups according to the angle of view.Secondly,the HOG（histograms of oriented gradients） and LBP（local binary pattern） features of each group were computed. In order to fast compute feature pyramids, a hierarchical feature extraction algorithm was proposed.And then latent SVM（support vector machine） was used to train the model parameters to obtain several components.Finally,these components were combined to form a mixture model.The model proposed was tested on FDDB and AFW databases, the results prove that proposed method can effectively detect multi-view faces in the wild, and outperform existing methods.