基于AAM的人脸图像年龄组分类研究

Research on AAM-based facial images age-group classification

针对随年龄的增长人脸图像年龄组分类准确率下降的问题,提出一种有效提高准确率和分类速度的年龄组分类方法。该方法结合主成分分析(PCA)和局部保持投影(LPP)建立主动外观模型(AAM),并对人脸图像进行特征点提取;在反向组合算法的基础上添加全局几何变换,增强AAM的表征能力,对输入人脸图像进行匹配;并利用支持向量机回归算法(SVR)对人脸图像进行年龄组分类。实验结果表明,该方法的年龄组分类准确率由79%提高到84%,分类耗时明显改善,且该方法更适用于亚洲人脸图像。

The new age-group classification with higher accuracy and speed is proposed to solve the problem of declining accu- racy of facial images age-group classification as the increasing age. Active appearance model （AAM） is built by combining principal component analysis with locality preserving projections, and the features of face images are extracted. The inverse composition algorithm is used to match face images and enhance the representational capacity of AAM by global geometric transform. Finally SVR algorithm is used for face images age-group classification. The results demonstrate that the accuracy of the new age classification based on AAM improves from 79 % to 84%, and the classification speed is higher. This method applies to Asian faces better.