基于局部统计几何特征的人脸性别分类方法

Facial gender classification based on local statistical geometric feature

为解决自动人脸性别分类问题,提出一种基于局部统计几何特征的性别分类方法。融合人脸图像特征点定位信息与人脸三维数据的几何信息,建立具有统计意义的局部统计几何特征,可以被认为是基于多模态人脸信息的方法。该方法具有标准的表达形式,方便学习算法建立分类器;融合几何信息,对图像中的噪声、光照和人脸上的化妆等具有一定的鲁棒性;考虑三维网格数据中面片数量不同带来的影响,对网格化精度具有一定鲁棒性。为验证其有效性,基于三维人脸公开数据库(FRGC2.0和BosphorusDB)进行相关分类实验,实验结果表明,该方法能够得到比较精确的分类结果。

For automatic gender classification based on facial data,agender classification method based on local statistical geometric feature was proposed.Facial landmarks from facial image and geometric information from 3 Dface model were combined to build the local statistical geometric feature,which was regarded as the multi-model facial data based method.The method defined regular representation for face data and it was convenient to be trained using learning method.It was robust to image noisy,light and cosmetics in face,as well as robust to different accuracy of triangular mesh.To evaluate the performances of the classification method,the public 3 Dfacial database FRGC2.0 and BosphorusDB were used to be the test data.The classification results show that the proposed classification method is effective.