基于三维人脸数据的性别分类

Gender Categorization Based on 3D Face Data

提出了一种新颖的基于视觉词典直方图的三维人脸特征,并将其应用到三维人脸数据中实现了人脸性别分类。视觉词典直方图特征提取分为两个阶段,训练阶段和识别阶段。在训练阶段中,先是通过Gabor滤波器提取出三维人脸数据中的局部纹理特性,组成滤波响应向量集合,然后通过聚类算法得到向量中心,即三维人脸视觉词汇;在识别阶段中,将滤波响应向量与三维人脸视觉词汇进行映射,得出对应的视觉词汇直方图向量,即基于视觉词典直方图的人脸特征。在获取人脸特征后,采用SVM分类器实现性别分类。实验表明,该特征在性别分类中可以获得比其他广泛采用的表象特征(如Gabor滤波器、LBP等)更好的分类性能,充分证明了该特征实现了三维人脸描述有效性和鲁棒性的统一。

The article proposes a novel histogram feature based on 3D facial visual codes, and successfully applies this algorithm to 3D face gender categorization. Extracting facial histogram feature based on visual codes can be divided into 2 stages, training stage and categorization stage, In the training stage, first the local texture characteristics in 3D facial data are extracted using Gabor filters to a form filter response vector set, and then vector centers, clustering algorithm. In the categorization stage, the namely, 3D facial visual codes are obtained filter response vectors are mapped to the visual through codes, and the mapping results can be represented as a histogram feature vector. Gender classification is achieved by using SVM classifier based on the obtained histogram feature vector. Experimental results illustrate that our proposed visual code histogram feature achieves better categorization performance than the widely used appearance-based facial features, such as Gabor filter and LBP features, which demonstrates its effectiveness and robustness for 3D face gender categorization.