CAS-PEAL大规模中国人脸图像数据库及其基本评测介绍

人脸图像数据库是人脸识别算法研究、开发、评测的基础，具有重要的意义．介绍了自行创建并已经部分共享的CAS-PEAL大规模中国人脸图像数据库及其基准测试结果．CAS-PEAL人脸图像数据库中包含了1040名中国人共99450幅头肩部图像．所有图像在专门的采集环境中采集，涵盖了姿态、表情、饰物和光照4种主要变化条件，部分人脸图像具有背景、距离和时间跨度的变化．目前该人脸图像数据库的标准训练、测试子库已经公开发布．与其他已经公开发布的人脸图像数据库相比，CAS-PEAL人脸图像数据库在人数、图像变化条件等方面具有综合优势，将对人脸识别算法的研究、评测产生积极的影响．同时，作为以东方人为主的人脸图像数据库，CAS-PEAL人脸图像数据库也使人脸识别算法在不同人种之间的比较成为可能，利于人脸识别算法在国内的实用化．还给出了两种基准人脸识别算法(Eigenface和Correlation)和两个著名商业系统在该人脸图像数据库上的测试结果，定量分析了该人脸图像数据库对于人脸识别算法的难易程度和人脸识别算法的发展现状．

CAS-PEAL: A Large-Scale Chinese Face Database and Some Primary Evaluations

As the basis of research, development and evaluation of face recognition algorithms, face image database is of great importance. In this paper, we introduce the construction and basic content of the CAS-PEAL, large-scale Chinese face database and some primary evaluation results based on the database. The CAS-PEAL face database consists of 99?450 facial images of 1?040 Chinese individuals. All the images in the database were collected in specially designed environment with four principal variations of pose, expression, accessory and lighting, as well as three other variations in terms of background, distance and aging. Currently, the standard training set and probe set of CAS-PEAL face database have been made publicly available for research purpose only on a case-by-case basis. Compared with other public face databases, CAS-PEAL excels in its large-scale and variation modes and is expected to have positive impact on the development and evaluation of face recognition algorithms. In addition, as an oriental face image database, CAS-PEAL makes possible the comparison of algorithms' performance between different ethnic groups and will benefit the application of face recognition technology in China. This paper also gives the evaluation results of two basic face recognition algorithms (Eigenface and Correlation) and two commercial systems, explains the difficulty of the database to the face recognition algorithms and analyses the current development status of face recognition technology.