摘要
针对人脸全局特征用于人脸验证存在的局限性,本文在Joint Bayesian人脸识别方法的基础上提出了基于局部贝叶斯分类器融合的人脸验证方法。该方法使用约束局部模型(CLM)在人脸上标注27个局部特征点,提取以这些特征点为中心的人脸块,并将它们进一步划分为互不重叠的若干个单元格;将这些人脸块的局部二值模式(LBP)特征通过Joint Bayesian统计训练得到多个局部分类器;最后利用逻辑回归模型将局部分类器融合为人脸验证分类器。在LFW(Labeled Face in the Wild)和WDRef(Wide and Deep Reference)数据库上进行了性能验证实验,实验结果表明该方法的性能要优于Joint Bayesian和其他现有典型分类器。
A novel face verification model based on confusing local Bayesian classifier will be proposed to eliminate the limitation of using global face feature for face verification. Firstly, 27 landmarks were located based on a Constrain Local Model（CLM） model. Then, face patches centered on each landmark were extracted and further split into non-overlapping cells. These face patches＇ Local Binary Pattern（LBP） feature can be used for creating local Bayesian classifiers by doing Joint Bayesian training. And the local classifiers were integrated in the framework of logistic regression. Finally, a face verification model was taken shape. The original approach was evaluated on the Labeled Face in the Wild（LFW） and Wide and Deep Reference（WDRef） databases. The experimental results show that our method is superior to Joint Bayesian method and most of the state-of-the-art classifiers.
出处
《光电工程》
CAS
CSCD
北大核心
2016年第3期80-87,共8页
Opto-Electronic Engineering
基金
国家自然科学基金资助项目(60972114)
关键词
人脸验证
LBP
贝叶斯分类器
分类器融合
逻辑回归
face verification
LBP
Bayesian classifier
classifier fusion
logistic regression
