基于连续Adaboost算法的多角度人脸检测技术研究与实现
|
摘要
随着数字图像处理技术以及智能学习算法的不断发展，人脸检测技术正越来越多的应用到视频监控、人机交互以及电子商务等领域；而所谓的人脸检测过程就是指从静态图像或动态视频帧图像中将人脸对象从背景范围中分割出来，并且指定人脸区域范围的过程。目前，成熟的人脸检测大部分是针对正脸规则样本进行检测，对于多角度的人脸检测，研究和实用的实例较少，因此研究如何进行高效、准确的多角度人脸检测成为研究人员越来越关注的焦点问题。
本文首先对已有人脸检测技术进行了总结和归纳，将其划分为基于特征的人脸检测方法、基于模板匹配的人脸检测方法和基于统计的人脸检测方法三类，并指出目前应用最为广泛、精度和效率最好的是基于统计的人脸检测中的基于积分图像特征的方法，由此引出本文的核心算法——Adaboost算法；其次，对Adaboost算法的相关应用基础技术——Haar特征及其扩展和积分图技术进行了深入研究，为运用Adaboost算法进行人脸检测打下基础；同时，还总结了连续Adaboost算法在人脸检测中的应用流程，并就连续Adaboost算法的相关改进工作的进展做了研究和阐述，指出基于查找表型弱分类器构建和基于多重阈值划分弱分类器构建两种弱分类器的改进方法，并指出它们在训练速度上存在缺陷。针对多角度人脸的检测场景，本文研究和实施了多角度人脸划分，将多角度人脸划分为84种，并通过研究Haar特征的特点，将需要通过样本学习而形成的分类器减少至12个，极大的提高了分类器的在训练时的速度；同时，由于整体项目的要求，对人脸检测的总体流程做了相应的改进，增加了图像类型检测、肤色检测、绝对位置检测和相对位置检测四个步骤，最终实现了一个基于连续Adaboost算法的多角度人脸检测系统(CAMFDS)。
本文所实现的基于连续Adaboost算法的多角度人脸检测系统(CAMFDS)采用MIT公开的人脸训练样本集做为训练样本，以2157张来自Internet的人脸图像作为测试样本集，针对彩色和灰度、闭眼和睁眼、正常和小眼以及有无眼镜等情形进行了分类测试，测试结果表明，人脸检测的准确率最低达到88.9%，召回率最低达到80.0%，执行效率统计每张图像总流程耗时708.7毫秒，其中包含预处理耗时、Adaboost算法执行耗时以及改进流程耗时。基于连续Adaboost算法的多角度人脸检测系统作为人体周边物体特征提取系统中的重要组成子系统，为整体系统提供了精确和高效的人脸定位，为后续检测模块的顺利实施打下了坚实的基础，人体周边物体特征提取系统目前已验收成功，整体项目已结项，与之相关的发明专利1项已公开。
With the continuous development of the digital image processing technology and intelligentlearning algorithms, Face Detection technology is increasingly being applied to video surveillance,human-computer interaction and e-commerce and other fields. Face detection is a process ofcarving out the face objects from the static images or dynamic video frame background, andspecifying the range of face area. Currently, the mature detection technologies are mostly used forthe frontal face detection, while the research and practical cases for multi-angle face detection areinfrequent. Therefore, how to efficiently and accurately detect the multi-angle face is increasinglybecoming the focus of researchers’attention.
Firstly, the paper summarizes the existed face detection technology, and divides it into threecategories: Feature-based face detection method, Based on template matching face detectionmethod and Statistics-based face detection method. The paper points out that, at present the mostwidely used, the best accuracy and efficiency method is based on image features in statistics-basedface detection method, and this paper introduces a core algorithm-Adaboost algorithm. Secondly,this paper carries on deep research about basic technology related to adaboost algorithm-Haarfeatures and extensions and integral image technology, for the use of adaboost face detectionalgorithm to lay the foundation; at the same time, generalizes the process of continuous adaboostface detection algorithm in the application. This paper researchs and explains the progress of theimprovement of the continuous adaboost algorithm. This paper also indicates that the Look-uptable-based weak classifiers building method and multiple threshold division-based weak classifiersbuilding method are two kinds of improvement methods of weak classifiers and demonstrates theirshortcomings in the training speed. For the multi-angle face detection application scene, this paperresearchs and implements multi-angle face division, divides the multi-angle face into 84 kinds; Atthe same time, by studying the characteristics of haar features, we reduce the classifiers to 12formed by sample study. As the overall project requirements, this paper improves the overallprocess of face detection, increases four steps in the process: Image type detection, Color detection,The absolute position detection and The relative position detection, and finally, achieves acontinuous adaboost algorithm-based multi-angle face detection system (CAMFDS).
CAMFDS uses MIT open face training set as training samples, 2157 face images from theInternet as a test sample set, and tests the classification of color and grayscale, eyes closed and eyesopen, normal and small eyes, with or without glasses cases. The result shows that face detection’s accuracy achieve 88.9% at least, recall rate achieve 80.0% at least, the total process of each imagetakes 708.7ms that includes the time-consuming of pre-processing, improved process and adaboostalgorithm execution. CAMFDS as an important component subsystem of Human peripheral objectfeature extraction system, provides a precise and efficient positioning of the human face, and lays asolid foundation for the successful implementation of the follow-up test modules. Human peripheralobject feature extraction system has been successfully checked and accepted. The overall projecthas been finished,and a related invention patent has been published.
本文首先对已有人脸检测技术进行了总结和归纳，将其划分为基于特征的人脸检测方法、基于模板匹配的人脸检测方法和基于统计的人脸检测方法三类，并指出目前应用最为广泛、精度和效率最好的是基于统计的人脸检测中的基于积分图像特征的方法，由此引出本文的核心算法——Adaboost算法；其次，对Adaboost算法的相关应用基础技术——Haar特征及其扩展和积分图技术进行了深入研究，为运用Adaboost算法进行人脸检测打下基础；同时，还总结了连续Adaboost算法在人脸检测中的应用流程，并就连续Adaboost算法的相关改进工作的进展做了研究和阐述，指出基于查找表型弱分类器构建和基于多重阈值划分弱分类器构建两种弱分类器的改进方法，并指出它们在训练速度上存在缺陷。针对多角度人脸的检测场景，本文研究和实施了多角度人脸划分，将多角度人脸划分为84种，并通过研究Haar特征的特点，将需要通过样本学习而形成的分类器减少至12个，极大的提高了分类器的在训练时的速度；同时，由于整体项目的要求，对人脸检测的总体流程做了相应的改进，增加了图像类型检测、肤色检测、绝对位置检测和相对位置检测四个步骤，最终实现了一个基于连续Adaboost算法的多角度人脸检测系统(CAMFDS)。
本文所实现的基于连续Adaboost算法的多角度人脸检测系统(CAMFDS)采用MIT公开的人脸训练样本集做为训练样本，以2157张来自Internet的人脸图像作为测试样本集，针对彩色和灰度、闭眼和睁眼、正常和小眼以及有无眼镜等情形进行了分类测试，测试结果表明，人脸检测的准确率最低达到88.9%，召回率最低达到80.0%，执行效率统计每张图像总流程耗时708.7毫秒，其中包含预处理耗时、Adaboost算法执行耗时以及改进流程耗时。基于连续Adaboost算法的多角度人脸检测系统作为人体周边物体特征提取系统中的重要组成子系统，为整体系统提供了精确和高效的人脸定位，为后续检测模块的顺利实施打下了坚实的基础，人体周边物体特征提取系统目前已验收成功，整体项目已结项，与之相关的发明专利1项已公开。
With the continuous development of the digital image processing technology and intelligentlearning algorithms, Face Detection technology is increasingly being applied to video surveillance,human-computer interaction and e-commerce and other fields. Face detection is a process ofcarving out the face objects from the static images or dynamic video frame background, andspecifying the range of face area. Currently, the mature detection technologies are mostly used forthe frontal face detection, while the research and practical cases for multi-angle face detection areinfrequent. Therefore, how to efficiently and accurately detect the multi-angle face is increasinglybecoming the focus of researchers’attention.
Firstly, the paper summarizes the existed face detection technology, and divides it into threecategories: Feature-based face detection method, Based on template matching face detectionmethod and Statistics-based face detection method. The paper points out that, at present the mostwidely used, the best accuracy and efficiency method is based on image features in statistics-basedface detection method, and this paper introduces a core algorithm-Adaboost algorithm. Secondly,this paper carries on deep research about basic technology related to adaboost algorithm-Haarfeatures and extensions and integral image technology, for the use of adaboost face detectionalgorithm to lay the foundation; at the same time, generalizes the process of continuous adaboostface detection algorithm in the application. This paper researchs and explains the progress of theimprovement of the continuous adaboost algorithm. This paper also indicates that the Look-uptable-based weak classifiers building method and multiple threshold division-based weak classifiersbuilding method are two kinds of improvement methods of weak classifiers and demonstrates theirshortcomings in the training speed. For the multi-angle face detection application scene, this paperresearchs and implements multi-angle face division, divides the multi-angle face into 84 kinds; Atthe same time, by studying the characteristics of haar features, we reduce the classifiers to 12formed by sample study. As the overall project requirements, this paper improves the overallprocess of face detection, increases four steps in the process: Image type detection, Color detection,The absolute position detection and The relative position detection, and finally, achieves acontinuous adaboost algorithm-based multi-angle face detection system (CAMFDS).
CAMFDS uses MIT open face training set as training samples, 2157 face images from theInternet as a test sample set, and tests the classification of color and grayscale, eyes closed and eyesopen, normal and small eyes, with or without glasses cases. The result shows that face detection’s accuracy achieve 88.9% at least, recall rate achieve 80.0% at least, the total process of each imagetakes 708.7ms that includes the time-consuming of pre-processing, improved process and adaboostalgorithm execution. CAMFDS as an important component subsystem of Human peripheral objectfeature extraction system, provides a precise and efficient positioning of the human face, and lays asolid foundation for the successful implementation of the follow-up test modules. Human peripheralobject feature extraction system has been successfully checked and accepted. The overall projecthas been finished,and a related invention patent has been published.
引文
[1]梁路宏，艾海舟，徐光祐等．人脸检测研究综述[J].计算机学报.2002，25(5):449－458.
[2]朱怀毅．人脸检测与识别的多特征分析与系统实现[D].上海:上海交通大学,2008.
[3] Paul Viola, Michael Jones. Rapid Object Detection using a Boosted Cascade of Simple Features[C]. Computer Vision and Pattern Recognition, 2001. CVPR 2001. Proceedings of the 2001IEEE Computer Society Conference on,2001,1:511-518.
[4] Colin Das, Aaron Becker, Timothy Bretl. Probably Approximately Correct Coverage for Robotswith Uncertainty[C]. 2011 IEEE/RSJ International Conference on Intelligent Robots andSystems, San Francisco,2011:1160-1166.
[5] Schapire R E.The Strength of Weak Learnability[J].Machine Learning,1990,5(2):197-227.
[6] Freund Y.Boosting a Weak Algorithm by Majority[J].Information and Computation,1995,121(2):256-285.
[7] Freund Y,Schapire R E.A Decision-Theoretic Generalization of On-Line Learning andAn Application to Boosting[J].Journal of Computer and System Sciences,1997,55(1):119-139.
[8]左登宇．基于Adaboost算法的人脸检测研究[D]．合肥：中国科学技术大学，2009.
[9] R. E. Schapire, and Y. Singer. Improved boosting algorithms using confidence-rated predictions.Machine Learning, 1999, 37(3): 297-336.
[10]武勃．多角度人脸检测与人口统计学分类[D]．北京：清华大学，2004．
[11]梁路宏，艾海舟，何克忠，张钹．基于仿射模板匹配的多角度单人脸定位[J].计算机学报．2000，23(6)：640－645．
[12]梁路宏，艾海舟，何克忠，张钹．基于多关联模板匹配的人脸检测[J]．软件学报．2001，12(1):94-102.
[13] Shan Du, Ward, R.K.. Adaptive Region-Based Image Enhancement Method for Robust FaceRecognition Under Variable Illumination Conditions. Circuits and Systems for VideoTechnology[J], IEEE Transactions on circuits and systems for video technology,2010,20(9):1165-1175.
[14] Zhibo Guo1,Yunyang Yan,Chunxia Zhao,Jingyu Yang. HTF-Boosting Learning and FaceDetection. 2008 IEEE Pacific-Asia Workshop on Computational Intelligence and IndustrialApplication[J].2008,1(19):376-380.
[15] Lee, K.-M., Li, Q., Daley, W.. Effects of Classification Methods on Color-Based FeatureDetection With Food Processing Applications. IEEE Transactions on Automation Science andEngineering[J].2007,4(1):40-51.
[16]艾海舟，肖习攀，徐光祐．人脸检测与检索[J]．计算机学报．2003，26(7)：874－881．
[17]亲柱晓，唐琏，郑群英．一种改进的人眼模板匹配算法[J]．计算机系统应用．2009，12：75－78．
[18]李智勇，田贞．基于肤色模型的人脸检测研究[J]．现代电子技术．2011，34(3)：131-133．
[19] Lee C H,Kim J S,Park K H. Automatic human face location in a complex background usingmotion and color information[J].Pattern Recognition.1996,29.11:1877-1899.
[20] Hilado,Samantha Denise F.,Dadios,Elmer P.,Gustilo,Reggie C.. Face Detection using NeuralNetworks with Skin Segmentation. Cybernetics and Intelligent Systems (CIS), 2011 IEEE 5thInternational Conference on.2011:261-265.
[21] C. Shavers, R. Li, and G. Lebby.An SVM-based Approach to Face Detection.2006 Proc. ofThirty-Eighth Southeastern Symp. On System Theory, pp. 362-366, Mar. 2006.
[22]王瑞平，陈杰，陈熙霖，高文．基于支持向量机的人脸检测训练集增强[J]．软件学报．2008，19(11)：2921－2931．
[23]王洪群，彭嘉雄，于秋则．采用动态HMM概率分布模型的人眼精确定位[J]．中国图像图形学报．2006，11(1)：26－32．
[24]谢倩茹，耿国华．复杂光照环境下视频人脸序列的自动检测方法[J]．计算机科学．2011，38(10)：267－269．
[25] Li, S.Z.,Zhenqiu Zhang. FloatBoost Learning and Statistical Face Detection. Pattern Analysisand Machine Intelligence, IEEE Transactions on.2004,26(9):1112-1123.
[26]朱文佳，戚飞虎．快速人脸检测与特征定位[J]．中国图像图形学报．2005，10(11)：1454-1458．
[27]张捷．基于核心灰度分布人脸自动检测方法研究[D]．郑州：中国人民解放军信息工程大学，2008．
[28]张洪明，赵德斌，高文．基于肤色模型、神经网络和人脸结构模型的平面旋转人脸检测[J]．计算机学报．2002，25(11)：1250－1256．
[29]钟志光．一种适用于人脸检测具有强聚类能力的新颜色空间YCH[J]．模式识别与人工智能．2011，4：45－51．
[30]王一伟．基于肤色和Adaboost算法的多角度多姿态人脸检测[D]．北京：北京交通大学，2011．
[31] Dal Y, Nakano Y. Face texture model based on SGLD and its application in face detection in acolor scene [J].Pattern Recognition, 1996,29(6):1007-1017.
[32]胡敏敏．人脸识别系统的研究与实现[D]．南京：南京邮电大学，2011．
[33]黎云汉．人脸特征提取与跟踪[D]．杭州：浙江大学，2008．
[34] ReisfeldD,YeshurunY,Robust deteetion offacial features by generalized symetry[C].Proeeedings of the International Conference on Pattern Reeognition.Hague.Netheriands.1992. 117-120.
[35] Zabrodsky H,Peleg S,Avnir D.Symmetry as a continuous feature.IEEE Transactions on PatternAnalysis and Machine Intelligence.1995.17(12).1154-1161.
[36] Saber E, Tekalp A M. Frontal- view face detection and facial feature extraction using color,shape, and symme try-based cost functions[J]. Pattern Recognition Letters, 1998, 19(8): 669-680.
[37] Leung T K, BurlM C, PeronaP. Finding faces in a cluttered scenes using random labeled graphmatching[C]. In Proceedings of the International Conference on Computer Vision, MIT,Cambridge, Massachusetts, USA, 1995, 637-644.
[38] King- Hsuan,Yang, David,J .Kriegman,Narendra Ahuja. Detecting Faces in Images: A Survey［J］.IEEE TRANSACTIONS ON PAMI,2002,VOL.24 .NO.1 JANUARY:34-58.
[39]刘翼光，沈理．利用Hausdroff距离人脸图像定位算法[J]．计算机研究与发展．2001，38 (4)：475－481．
[40]梁路宏，艾海舟，肖习攀等．基于模板匹配与支持矢量机的人脸检测[J]．计算机学报．2002，25(1)：22－29．
[41] YuilleA L, H a llinan PW, Cohen D S. Feature extraction from faces using deformabletemplates[J]. International Journal of Computer Vision, 1992, 8(2): 99-111.
[42] Lanitis A, Taylor C J, Cootes T F. An automatic face identification system using flexibleappearance models[J]. Image and Vision Computing, 1995, 13(5): 393- 402.
[43]柳莲花．视频监控系统的人脸检测与识别技术研究[D]．南京：南京航空航天大学，2010．
[44] Rowley H A,Baluja S,Kanade T.Neural network-based face detection[J].IEEE Trans.PatternAnal.Mach.Intell.20,January 1998:23-38.
[45] Vapnik V.The Nature of Statistical Learning Theory[M].New York:Springer-Verlag,1995.
[46] Osuna E, Freund R, Girosi R. Training support vector machines[C].An application to facedetection,in IEEE Proc.of Int.Conf.on Computer Vision and Pattern Recognition, 6,1997.
[47] Heiselet, B.,Serre, T.,Pontil, M.,Poggio, T. . Component-based Face Detection. ComputerVision and Pattern Recognition[C], 2001. CVPR 2001. Proceedings of the 2001 IEEEComputer Society Conference on.2001,1:657-662.
[48] Nefian A V,Hayes M H.Face detection and recognition using hidden markov models[C].Proc.IEEE Conf.on Image Processing,Chicago,1998:141-145.
[49]谢红跃，方星春，蔡起运．一种新的改进Adaboost弱分类器训练算法[J]．中国图像图形学报．2009，14(11)：2411－2415．
[50]李闯，丁晓青，吴佑寿．一种改进的AdaBoost算法——AD AdaBoost[J]．计算机学报．2007，30(1)：103－109．
[51] Rainer Lienhart,Jochen Maydt.An Extended Set of Harr-like Features for Rapid ObiectDetection[C].IEEE ICIP 2002,VOL.1:900-903.
[52]刘晓克，孙燮华，周永霞．基于新Haar-like特征的多角度人脸检测[J]．计算机工程．2009，35(19)：195－197．
[53]武勃，黄畅．基于连续Adaboost算法的多视角人脸检测[J]．计算机研究与发展．2005，42(9)：1612－1621．
[54]孙士明，潘青，纪友芳．多阈值划分的连续AdaBoost人脸检测[J]．计算机应用．2009，29(8)：1098－2104．
[2]朱怀毅．人脸检测与识别的多特征分析与系统实现[D].上海:上海交通大学,2008.
[3] Paul Viola, Michael Jones. Rapid Object Detection using a Boosted Cascade of Simple Features[C]. Computer Vision and Pattern Recognition, 2001. CVPR 2001. Proceedings of the 2001IEEE Computer Society Conference on,2001,1:511-518.
[4] Colin Das, Aaron Becker, Timothy Bretl. Probably Approximately Correct Coverage for Robotswith Uncertainty[C]. 2011 IEEE/RSJ International Conference on Intelligent Robots andSystems, San Francisco,2011:1160-1166.
[5] Schapire R E.The Strength of Weak Learnability[J].Machine Learning,1990,5(2):197-227.
[6] Freund Y.Boosting a Weak Algorithm by Majority[J].Information and Computation,1995,121(2):256-285.
[7] Freund Y,Schapire R E.A Decision-Theoretic Generalization of On-Line Learning andAn Application to Boosting[J].Journal of Computer and System Sciences,1997,55(1):119-139.
[8]左登宇．基于Adaboost算法的人脸检测研究[D]．合肥：中国科学技术大学，2009.
[9] R. E. Schapire, and Y. Singer. Improved boosting algorithms using confidence-rated predictions.Machine Learning, 1999, 37(3): 297-336.
[10]武勃．多角度人脸检测与人口统计学分类[D]．北京：清华大学，2004．
[11]梁路宏，艾海舟，何克忠，张钹．基于仿射模板匹配的多角度单人脸定位[J].计算机学报．2000，23(6)：640－645．
[12]梁路宏，艾海舟，何克忠，张钹．基于多关联模板匹配的人脸检测[J]．软件学报．2001，12(1):94-102.
[13] Shan Du, Ward, R.K.. Adaptive Region-Based Image Enhancement Method for Robust FaceRecognition Under Variable Illumination Conditions. Circuits and Systems for VideoTechnology[J], IEEE Transactions on circuits and systems for video technology,2010,20(9):1165-1175.
[14] Zhibo Guo1,Yunyang Yan,Chunxia Zhao,Jingyu Yang. HTF-Boosting Learning and FaceDetection. 2008 IEEE Pacific-Asia Workshop on Computational Intelligence and IndustrialApplication[J].2008,1(19):376-380.
[15] Lee, K.-M., Li, Q., Daley, W.. Effects of Classification Methods on Color-Based FeatureDetection With Food Processing Applications. IEEE Transactions on Automation Science andEngineering[J].2007,4(1):40-51.
[16]艾海舟，肖习攀，徐光祐．人脸检测与检索[J]．计算机学报．2003，26(7)：874－881．
[17]亲柱晓，唐琏，郑群英．一种改进的人眼模板匹配算法[J]．计算机系统应用．2009，12：75－78．
[18]李智勇，田贞．基于肤色模型的人脸检测研究[J]．现代电子技术．2011，34(3)：131-133．
[19] Lee C H,Kim J S,Park K H. Automatic human face location in a complex background usingmotion and color information[J].Pattern Recognition.1996,29.11:1877-1899.
[20] Hilado,Samantha Denise F.,Dadios,Elmer P.,Gustilo,Reggie C.. Face Detection using NeuralNetworks with Skin Segmentation. Cybernetics and Intelligent Systems (CIS), 2011 IEEE 5thInternational Conference on.2011:261-265.
[21] C. Shavers, R. Li, and G. Lebby.An SVM-based Approach to Face Detection.2006 Proc. ofThirty-Eighth Southeastern Symp. On System Theory, pp. 362-366, Mar. 2006.
[22]王瑞平，陈杰，陈熙霖，高文．基于支持向量机的人脸检测训练集增强[J]．软件学报．2008，19(11)：2921－2931．
[23]王洪群，彭嘉雄，于秋则．采用动态HMM概率分布模型的人眼精确定位[J]．中国图像图形学报．2006，11(1)：26－32．
[24]谢倩茹，耿国华．复杂光照环境下视频人脸序列的自动检测方法[J]．计算机科学．2011，38(10)：267－269．
[25] Li, S.Z.,Zhenqiu Zhang. FloatBoost Learning and Statistical Face Detection. Pattern Analysisand Machine Intelligence, IEEE Transactions on.2004,26(9):1112-1123.
[26]朱文佳，戚飞虎．快速人脸检测与特征定位[J]．中国图像图形学报．2005，10(11)：1454-1458．
[27]张捷．基于核心灰度分布人脸自动检测方法研究[D]．郑州：中国人民解放军信息工程大学，2008．
[28]张洪明，赵德斌，高文．基于肤色模型、神经网络和人脸结构模型的平面旋转人脸检测[J]．计算机学报．2002，25(11)：1250－1256．
[29]钟志光．一种适用于人脸检测具有强聚类能力的新颜色空间YCH[J]．模式识别与人工智能．2011，4：45－51．
[30]王一伟．基于肤色和Adaboost算法的多角度多姿态人脸检测[D]．北京：北京交通大学，2011．
[31] Dal Y, Nakano Y. Face texture model based on SGLD and its application in face detection in acolor scene [J].Pattern Recognition, 1996,29(6):1007-1017.
[32]胡敏敏．人脸识别系统的研究与实现[D]．南京：南京邮电大学，2011．
[33]黎云汉．人脸特征提取与跟踪[D]．杭州：浙江大学，2008．
[34] ReisfeldD,YeshurunY,Robust deteetion offacial features by generalized symetry[C].Proeeedings of the International Conference on Pattern Reeognition.Hague.Netheriands.1992. 117-120.
[35] Zabrodsky H,Peleg S,Avnir D.Symmetry as a continuous feature.IEEE Transactions on PatternAnalysis and Machine Intelligence.1995.17(12).1154-1161.
[36] Saber E, Tekalp A M. Frontal- view face detection and facial feature extraction using color,shape, and symme try-based cost functions[J]. Pattern Recognition Letters, 1998, 19(8): 669-680.
[37] Leung T K, BurlM C, PeronaP. Finding faces in a cluttered scenes using random labeled graphmatching[C]. In Proceedings of the International Conference on Computer Vision, MIT,Cambridge, Massachusetts, USA, 1995, 637-644.
[38] King- Hsuan,Yang, David,J .Kriegman,Narendra Ahuja. Detecting Faces in Images: A Survey［J］.IEEE TRANSACTIONS ON PAMI,2002,VOL.24 .NO.1 JANUARY:34-58.
[39]刘翼光，沈理．利用Hausdroff距离人脸图像定位算法[J]．计算机研究与发展．2001，38 (4)：475－481．
[40]梁路宏，艾海舟，肖习攀等．基于模板匹配与支持矢量机的人脸检测[J]．计算机学报．2002，25(1)：22－29．
[41] YuilleA L, H a llinan PW, Cohen D S. Feature extraction from faces using deformabletemplates[J]. International Journal of Computer Vision, 1992, 8(2): 99-111.
[42] Lanitis A, Taylor C J, Cootes T F. An automatic face identification system using flexibleappearance models[J]. Image and Vision Computing, 1995, 13(5): 393- 402.
[43]柳莲花．视频监控系统的人脸检测与识别技术研究[D]．南京：南京航空航天大学，2010．
[44] Rowley H A,Baluja S,Kanade T.Neural network-based face detection[J].IEEE Trans.PatternAnal.Mach.Intell.20,January 1998:23-38.
[45] Vapnik V.The Nature of Statistical Learning Theory[M].New York:Springer-Verlag,1995.
[46] Osuna E, Freund R, Girosi R. Training support vector machines[C].An application to facedetection,in IEEE Proc.of Int.Conf.on Computer Vision and Pattern Recognition, 6,1997.
[47] Heiselet, B.,Serre, T.,Pontil, M.,Poggio, T. . Component-based Face Detection. ComputerVision and Pattern Recognition[C], 2001. CVPR 2001. Proceedings of the 2001 IEEEComputer Society Conference on.2001,1:657-662.
[48] Nefian A V,Hayes M H.Face detection and recognition using hidden markov models[C].Proc.IEEE Conf.on Image Processing,Chicago,1998:141-145.
[49]谢红跃，方星春，蔡起运．一种新的改进Adaboost弱分类器训练算法[J]．中国图像图形学报．2009，14(11)：2411－2415．
[50]李闯，丁晓青，吴佑寿．一种改进的AdaBoost算法——AD AdaBoost[J]．计算机学报．2007，30(1)：103－109．
[51] Rainer Lienhart,Jochen Maydt.An Extended Set of Harr-like Features for Rapid ObiectDetection[C].IEEE ICIP 2002,VOL.1:900-903.
[52]刘晓克，孙燮华，周永霞．基于新Haar-like特征的多角度人脸检测[J]．计算机工程．2009，35(19)：195－197．
[53]武勃，黄畅．基于连续Adaboost算法的多视角人脸检测[J]．计算机研究与发展．2005，42(9)：1612－1621．
[54]孙士明，潘青，纪友芳．多阈值划分的连续AdaBoost人脸检测[J]．计算机应用．2009，29(8)：1098－2104．