人脸年龄估计和年龄面貌合成技术研究进展  

Research Progress on Human Facial Age Estimation and Synthesis of Age-correlated Appearance

作　　者：潘思宇[1] 赵雯婷[2] 唐鲲[3] 马新[2] 叶健[1,2] 李彩霞[2] PAN Siyu;ZHAO Wenting;TANG Kun;MA Xin;YE Jian;LI Caixia(People’s Public Security University of China, Beijing 100038;Institute of Forensic Science, Ministry of Public Security, National Engineering Laboratory for Forensic Science, Beijing Engineering Research Center of Crime Scene Evidence Examination, Beijing 100038, China;Shanghai Institutes for Biological Sciences, Chinese Academy of Science, Shanghai 200031, China)

机构地区：[1]中国人民公安大学,北京100038 [2]公安部物证鉴定中心现场物证溯源技术国家工程实验室北京市现场物证检验工程技术研究中心,北京100038 [3]中科院上海生命科学研究院,上海200031

出　　处：《刑事技术》2017年第4期270-276,共7页Forensic Science and Technology

基　　金：国家重点研发计划(No.2017YFC080351);中央级公益性科研院所基本科研业务费专项资金项目(No.2016JB001;No.2017JB025);公安部技术研究计划项目(No.2016JSYJC15)

摘　　要：近年来,由于在刑事侦查、失踪人口追踪、治安管理监测等领域巨大的应用价值,基于人脸图像的年龄估计和年龄面貌合成技术受到了广泛的关注。年龄估计是通过计算机技术给出脸部图像的确切年龄或年龄范围。年龄面貌合成则是根据脸部特征的年龄变化规律,基于现有的脸部图像推测过去或未来某个年龄的脸部形态。通过各种统计学及图像学分析方法,这两项技术在过去几十年间得到了迅速的发展。本文全面综述了目前在年龄估计及年龄面貌合成方面的技术进展,包括已有模型、主要和最新算法、系统模式等,分别比较不同模型及算法的优缺点,并对当前存在的技术难题及将来的发展方向进行了讨论。Human facial age estimation and aging progression are being increasingly paid much attention because of their huge value in the aspects of forensic investigation,missing people searching,security control and surveillance.Computer-based age estimation is defined to label a face image with an exact age or age scope.Age-correlated appearance synthesis(aging progression)is a process of re-rendering the images of an individual's face to represent the effect of aging or rejuvenating on their appearance based on the rule of facial features changing along with aging.Both techniques above have rapidly developed in the last decades thanks to the combination of the various ever-improving statistics and iconography.Facial aging correlates to some unique characteristics such as smoothness,face shape changes,wrinkles,drooping eyelids and pouches under the eyes,all of which make age estimation a challenging and complicated task.A large number of statistical and/or graphical models have been applied to promote the age estimation,including active appearance model(AAM),aging pattern subspace model(AGES),age manifold model and bio-inspired features(BIF)model.Most of these automatic models are based on classification and/or regression algorithms.Moreover,in order to find a particular progressive aging pattern and render aging faces in a personalized way,researchers have ameliorated the automatic appearance models by learning a set of age-group specific dictionaries or aging markers.In this review,the techniques were surveyed on facial image-based age estimation and age-correlated appearance synthesis:existing models,popular algorithms and their systematic performances.The technical difficulties were also discussed along with the future directions.

关 键 词：人脸图像 年龄估计 年龄面貌合成 

分 类 号：DF795.2[医药卫生—法医学]