摘要
提出了一种基于Gabor小波和局域二值模式(Local binary pattern,LBP)直方图序列的人脸年龄估计方法。首先对人脸图像提取多方向与多尺度的Gabor幅值域图谱(Gabor magnitude maps,GMMs);然后采用基于局部特征的LBP算子对GMMs编码,并对之分块,由各子块的直方图序列来描述人脸;为进一步降低人脸特征维数,再对人脸直方图序列特征应用主成分分析(PCA);最后使用支持向量机回归(SVR)的LOPO策略对人脸年龄库进行训练和测试。实验结果表明,该方法可以较为快速有效地对人脸图像进行年龄估计。
A method for age estimation of facial images is proposed based on the combination of the Gabor wavelets and the histogram sequence of the local binary pattern （LBP）. The facial images are firstly filtered by the multi-orientation and multi-scale Gabor before Gabor magnitude maps （GMMs） are extracted. Then the local neighbor pattern on GMMs is extracted by LBP based on local characteristics and the characteristics are divided into several sub-blocks to calculate the histogram sequences. To further reduce the dimension of facial features, Principal component analysis （PCA） is applied to the histogram sequences. Finally, a leave-one-personout （LOPO） test scheme of the support vector regression （SVR） is used to train and test the face age database. Experimental results show that the method can estimate the age of human faces quickly and effectively.
出处
《数据采集与处理》
CSCD
北大核心
2012年第3期340-345,共6页
Journal of Data Acquisition and Processing
基金
天津市科技支撑计划(10ZCKFGX00700)资助项目
关键词
支持向量机回归
局域二值模式
主成分分析
年龄估计
support vector regression（SVR）
local analysis（PCA）
age estimation binary pattern（LBP）
principal component
