基于实时预测学习分类的脸像快速检测方法  

Fast face detection method based on real-time prediction and learning classification

作　　者：刘长[1] 王雪[1] 张星[1] 

机构地区：[1]清华大学精密仪器与机械学系精密测试技术及仪器国家重点实验室,北京100084

出　　处：《仪器仪表学报》2012年第1期36-41,共6页Chinese Journal of Scientific Instrument

基　　金：国家重点基础研究发展计划(973计划)基金(2006CB303000);国家自然科学基金(60970103;60673176;60373014;50175056);教育部博士点基金(20090002110016)资助项目

摘　　要：视频图像中脸像检测是近年来视觉图像检测和模式识别领域的研究热点。提出一种基于实时预测学习分类的脸像快速检测算法,即ARMA-Boost算法。首先根据脸像位置先验信息,利用ARMA模型(auto-regressive and moving average model)预测脸像位置区域,然后采用AdaBoost算法对预测区域进行脸像检测。该方法在时间维度对AdaBoost算法进行扩展,减小脸像搜索范围,提高检测效率。利用该方法对离线视频文件和CCD图像传感器实时脸像视频进行检测,实验结果表明,与支持向量机、传统AdaBoost和基于优化肤色模型的AdaBoost改进算法相比,ARMA-Boost算法脸像检测准确率高,实时性更好,可以对视频脸像进行快速检测应用。Face detection is a popular research topic in computer vision and pattern recognition in resent years. This paper proposes a new fast face detection method, ARMA-Boost algorithm, based on real-time prediction and learning classification. During the detection process, ARMA model is first used to predict potential face areas. Then Ada- Boost algorithm searches faces in those areas. Experiment results of off-line video data and CCD real-time data indicate that compared with SVM, traditional AdaBoost and improved AdaBoost algorithms, the proposed ARMA-Boost algorithm can reduce detection time and also improve the accurate rate of face detection.

关 键 词：脸像快速检测 ARMA模型 实时预测 学习分类 ADABOOST算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]