摘要： 提出了一种基于神经网络和层次支持向量机的多姿态人脸识别方法。该方法在训练阶段先利用神经网络把姿态人脸图像特征向准标准人脸图像特征映射，再根据聚类结果来训练支持向量机。识别阶段是利用神经网络变换得到待识别图像所对应的准标准图像的特征，再让层次支持向量机初步判断待识别图像最可能所属的人，最后利用否定算法对待识别的人脸图像进行确认。实验表明该算法效果较佳。

关键词: 神经网络, 层次支持向量机, 离散余弦变换, 聚类算法

Abstract: The paper presents a method of pose-varied face recognition based on neural network and hierarchical support vector machines. At the stage of training, it transforms the feature vector of pose-varied image to the feature vector of standard image using neural network, then clusters the standard image feature vector, and trains the hierarchical support vector machines using the result of clustering. At the stage of the recognition, it transforms the feature vector of pose-varied image to the feature vector of standard image using the neural network, estimates which person the image mostly belongs to using hierarchical support vector machines, and confirms estimation using negative algorithm. The experiment shows that the effect is better.

Key words: Neural network, Hierarchical support vector machines, Discrete cosine transformation, Clustering algorithm