基于人脸图像的年龄估计  

Age Estimation Based on Facial Image

作　　者：林时苗[1] 毛晓蛟[1] 杨育彬[1] 

机构地区：[1]南京大学计算机软件新技术国家重点实验室,南京210093

出　　处：《计算机科学》2015年第6期32-36,共5页Computer Science

基　　金：教育部新世纪优秀人才计划(NCET-11-0213);国家自然科学基金(61273257;61321491;61035003);江苏省六大人才高峰项目(2013-XXRJ-018)资助

摘　　要：年龄是人固有的生物特征,随着年龄的变化,人脸特征也不断变化。近年来基于人脸图像的年龄估计方法的研究不断深入。基于人脸图像的年龄估计主要有两个阶段:特征提取和估计方法。针对以上两个阶段,分别提出相应的方法。在特征提取方面,为了更好地描述年龄变化,特别是针对未成年人,引入了方向梯度直方图(Histogram of Oriented Gradient,HOG)特征,并将其与局部二元模式(Local Binary Pattern,LBP)特征进行融合;在估计方法方面,提出了软双层估计模型,其采用由粗到细的策略。首先,在第一层将人脸分成"未成年人"与"成年人"两类;然后,在第二层通过在两类的边界设置重叠区域,分别对其建立年龄估计模型,以对第一层的错误分类进行补救。通过实验发现,融合的特征具有更强的年龄判别性,同时,软双层模型也进一步提高了年龄估计的准确度。Age is an inherent biometric for human. As we grow older,our faces will change a lot. Age estimation based on facial image has been widely studied in recent years. Age estimation mainly consists of two phases., feature extraction and estimation method. A new age estimation method was proposed in this paper. In the feature extraction phase, we suggested combining histogram of oriented gradient （HOG） with local binary patter （LBP） to better describe the age progression of facial images, especially for the teenagers. In the estimation phase, a soft two-level estimation method based on coarse-to-fine strategy was proposed. Specifically, facial images are categorized as either adults or teenagers in the first level. In the second level, then age estimation models for each of the categories are trained, and an overlap area at the category boundary is adopted to fix the classification errors caused by the first level. Experimental results show that the features of fusion achieve better discriminative power of aging. Moreover, the soft two-level model further im- proves the age estimation accuracy.

关 键 词：年龄估计 特征融合 软双层模型 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]