基于半监督回归学习的人脸几何美丽分数  

FACIAL GEOMETRIC BEAUTY SCORE BASED ON SEMI-SUPERVISED REGRESSION LEARNING

作　　者：戴礼青 金忠[1] 孙明明[1] 

机构地区：[1]南京理工大学计算机科学与工程学院,江苏南京210094

出　　处：《计算机应用与软件》2015年第3期209-211,219,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(61005005;61233011)

摘　　要：基于人脸美学的迅速发展,对人脸的几何特征定义、几何特征规范化以及几何特征对判断人脸美与否的贡献进行研究。首先定义人脸几何美丽分数函数,然后将流形学习与半监督学习相结合,用流形上的半监督回归方法学习人脸几何美丽分数。为了突出几何特征,还验证了人脸表情与几何美丽分数之间的关系。与K近邻(KNN)、支持向量机(SVM)、C4.5决策树分类方法相比,通过实验验证,证明了所提方法的有效性和可行性。Based on rapid development of facial aesthetics,we mainly study the definition of facial geometric feature,the normalisation of geometric features and the contribution of geometric features to judging whether the face is beauty or not. First,we define the facial geometric beauty score function,and then combine the manifold learning with semi-supervised learning,use semi-supervised regression on manifolds to learn geometric beauty score of faces. In order to highlight the geometric features,we also verify the relationship between facial expression and geometric beauty scores. Compared with KNN,SVM and C4. 5 decision tree classification methods,the validity and the feasibility of the proposed methods are proved by experiment.

关 键 词：人脸美学 人脸几何特征 几何美丽分数 流形学习 半监督回归 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]