人脸定位方法研究
|
摘要
人脸定位(face detection)是指在输入图像中确定所有人脸(如果存在)大小、位置、位姿的过程。人脸定位作为人脸信息处理中的一项关键技术,近年来成为模式识别与计算机视觉领域内一项受到普遍重视、研究十分活跃的课题。近年来人脸定位技术的研究已经取得了长足的进展,涌现出了许多新的人脸定位方法。目前人脸定位的主要方法分为四类:基于知识的方法、基于特征的方法、基于模板匹配的方法、基于外观学习的方法。在众多的人脸定位算法中,2001年Paul Viola和Michael Jones提出的基于AdaBoost算法的人脸定位方法从根本上解决了人脸定位的速度问题,同时具有较好的检测效果。
肤色是人脸的重要信息,不依赖面部的细节特征,对于旋转、表情等变化情况都能适用,具有相对的稳定性并且和大多数背景物体的颜色相区别,但是当背景复杂的情况下可能存在大量的类肤色物体时,导致了该算法较高的误检率。基于Adaboost的人脸定位其基本思想是当分类器对某一样本分类错误时,增加其权重;当分类正确时,则减少该样本的权重,使得后续的分类器更加强化对分类错误的样本进行训练,最终得到一个十分理想的分类器。该方法从根本上解决了人脸定位的速度问题,而且具有较低的误检率,但是该方法的检测率不如肤色检测的检测率高。
基于上述原因,本文采用肤色分割与AdaBoost相结合的方法进行人脸定位,在提高检测率的同时降低了误检率。本文首先,输入待检测图像,将其转换至YCbCr色彩空间,根据肤色的聚集范围将其二值化,并经过形态学预处理得到人脸的候选区域。在预处理过程中,采用可调节结构元素,解决了对于不同图像中人脸大小不一采用固定的结构元素造成的人脸丢失问题,提高了检测率。然后,将肤色分割后的人脸候选区域做为AdaBoost的输入窗口进行人脸定位。最后,在待检测图像中标记所得到的人脸区域。
Face detection (face detection) is the process that determine all of the face (if it exists) the size, location, position and orientation from input image. Face detection is a key technology of human face information processing, it becomes a attracted universal attention, very active research topic in the field of pattern recognition and computer vision. The research in face detection technology in recent years, has made considerable progress, there have been many new face detection method. Currently the primary method of face detection is divided into four categories: the method based on knowledge, the method based on characteristics, the method based on template matching, the method based on appearance and learning. Among the human face detection algorithm, Paul Viola and Michael Jones proposed algorithm based on AdaBoost face detection method in 2001. The method solved the speed of face detection problem in fundamentally, ane has good detection results.
Skin color is the important information of the human face, and it does not rely on the details of facial features, and the rotation, expression changing can be applied. It has relative stable and distinguish with most of the background objects.But the algorithm has led to a high false positive rate when the background is very complicated , there may be a large number of class color object case. The basic idea of based on AdaBoost face detection is that when the classifier to a sample of classification error, increase its weight; When the classification accuracy, then reduce the weight of the sample, making the classifier more intensive follow-up of the classification error of the sample training and eventually get a very good classifier. This method is a fundamental solution to the speed of face detection, but also has a lower false detection rate, end the method of detection rate is lower than skin color detection rate.
For above these reasons, in this paper, skin color segmentation method and combination of AdaBoost face detection was used. It improves the detection rate while reducing the false detection rate. In this paper, input the detecting image and convert it into YCbCr color space, make it to binary image according to the aggregation range of color values, and get candidate face region after morphological preprocessing. In the pretreatment process, the using adjusted structural elements can solve human face loss problem bacause of the human face images of different sizes in different image uesd the stable structural elements and improved the detection rate. Then, after the face color segmentation candidate region as the input windows of AdaBoost face detection. Finally, in the detection of an image tag to be received by the face region.
肤色是人脸的重要信息,不依赖面部的细节特征,对于旋转、表情等变化情况都能适用,具有相对的稳定性并且和大多数背景物体的颜色相区别,但是当背景复杂的情况下可能存在大量的类肤色物体时,导致了该算法较高的误检率。基于Adaboost的人脸定位其基本思想是当分类器对某一样本分类错误时,增加其权重;当分类正确时,则减少该样本的权重,使得后续的分类器更加强化对分类错误的样本进行训练,最终得到一个十分理想的分类器。该方法从根本上解决了人脸定位的速度问题,而且具有较低的误检率,但是该方法的检测率不如肤色检测的检测率高。
基于上述原因,本文采用肤色分割与AdaBoost相结合的方法进行人脸定位,在提高检测率的同时降低了误检率。本文首先,输入待检测图像,将其转换至YCbCr色彩空间,根据肤色的聚集范围将其二值化,并经过形态学预处理得到人脸的候选区域。在预处理过程中,采用可调节结构元素,解决了对于不同图像中人脸大小不一采用固定的结构元素造成的人脸丢失问题,提高了检测率。然后,将肤色分割后的人脸候选区域做为AdaBoost的输入窗口进行人脸定位。最后,在待检测图像中标记所得到的人脸区域。
Face detection (face detection) is the process that determine all of the face (if it exists) the size, location, position and orientation from input image. Face detection is a key technology of human face information processing, it becomes a attracted universal attention, very active research topic in the field of pattern recognition and computer vision. The research in face detection technology in recent years, has made considerable progress, there have been many new face detection method. Currently the primary method of face detection is divided into four categories: the method based on knowledge, the method based on characteristics, the method based on template matching, the method based on appearance and learning. Among the human face detection algorithm, Paul Viola and Michael Jones proposed algorithm based on AdaBoost face detection method in 2001. The method solved the speed of face detection problem in fundamentally, ane has good detection results.
Skin color is the important information of the human face, and it does not rely on the details of facial features, and the rotation, expression changing can be applied. It has relative stable and distinguish with most of the background objects.But the algorithm has led to a high false positive rate when the background is very complicated , there may be a large number of class color object case. The basic idea of based on AdaBoost face detection is that when the classifier to a sample of classification error, increase its weight; When the classification accuracy, then reduce the weight of the sample, making the classifier more intensive follow-up of the classification error of the sample training and eventually get a very good classifier. This method is a fundamental solution to the speed of face detection, but also has a lower false detection rate, end the method of detection rate is lower than skin color detection rate.
For above these reasons, in this paper, skin color segmentation method and combination of AdaBoost face detection was used. It improves the detection rate while reducing the false detection rate. In this paper, input the detecting image and convert it into YCbCr color space, make it to binary image according to the aggregation range of color values, and get candidate face region after morphological preprocessing. In the pretreatment process, the using adjusted structural elements can solve human face loss problem bacause of the human face images of different sizes in different image uesd the stable structural elements and improved the detection rate. Then, after the face color segmentation candidate region as the input windows of AdaBoost face detection. Finally, in the detection of an image tag to be received by the face region.
引文
[1] Yang M H, Ahuja N, Kriegman D. A survey on face detection methods. 1999, http://vision.aiuiuc. edu/mhyang/papers/survey.ps.gz.
[2]梁路宏,艾海舟,徐光佑等.人脸定位研究综述.计算机学报,2002,5(25):449-459.
[3]周杰,卢春雨,张长水等.人脸自动识别方法综述.电子学报,2000,28(4):102-106.
[4]贾云得.机器视觉.北京:科学出版社,2000.
[5]江嘉明.基于自然图片的人脸定位识别系统的实现:(硕士学位论文).大连:大连理工大学,2008,6.
[6] Brunelli R,Poggiot T.Template matching method resarch.. New York: Academic Pr,1999.
[7] Lee CH,KimJS,Park KH. Automatic human face location in a complex background using motion and color information.Pattern Recognition,1996,29(11):1877-1899.
[8] DalY, NakanoY.Face texture model based on SGLD and application in face detection in a color scene. Pattern Recognition,1996,29(6):1007-1017.
[9] Moghaddam B,Pentland A. Probabilistic Visual Learning for Object Representation. IEEE Transon Pattern Analysis and Machine Intelligence,1997,19(7):696-710.
[10] Rowley HA, Baluja S, Kanade T. Rotation invariant neural network-based face detection, CM U-CS-97-201, 1997.
[11] Saber E,Tekalp AM.Frontal-view face detection and facial feature extraction using color shape and symmetry based cost functions[J]. Pattern Recognition Letters, 1998, 19(8): 669-680.
[12]邓刚,闫胜业,张洪明。人脸定位技术报告,2001.
[13]罗敏.基于肤色的人脸定位和面部特征定位技术研究:(硕士学位论文).江苏:江苏大学,2008,5.
[14]唐伟,陈兆乾,吴建鑫等.静态灰度图像中的人脸定位方法综述.计算机科学,2002,29(2):134-137.
[15] Yang G, Huang T.s.Human face detection in complex background.Pattern Recognition. 1994,27(1):53-63.
[16] Kotropulos C., Pitas I. Rule-based face detection in frontal viwes. In Proceedings of International Conference on Acoustics, Speech and Singal Processing, 1996, 4:312-317.
[17] DaiY., NakanoY, Extraction for facial images from complex background using color information and SGLD matrices.In Proceeding of the First International Workshop on Automatic Face and Gesture Recognition,1997,(4):2537-2540.
[18] Turk M., Pentland A.Eingenface for recognition .Journal of congnitive neuroscience, 1991, (1): 71-86.
[19] Pentland A, MoghaddamB, StarnerT.View_based and modular eigenspaces for face recongnition. Seattle, WA, 1994.21-23.
[20] Osuna E, Freund R, Girosi F.Training support vector machines: an application to face detection. Proceddings of Computer Vision and Pattern Recongnition Puerto Rico, 1997.130-136.
[21] Viola P, Jones M.Robust real time objects detection. International Journal of Computer Vision 2002, 57(2):13-7154.
[22] Hsu.R.L, Mohamed Abdel-Mottaleb M, Jain A.K. Face Detection in Color Images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24(5): 696-706.
[23] Yow K.C, Cipolla R.Feauter-based human face detection. Image and Vision Computing, 1997, 15(9): 713-735.
[24] YullieA, HallinanP, CohenD. Feature exaction from faces using deformable templates. Internation al Journal of Computing Vision, 1992, 8(2): 99-111.
[25] Miao J, YinB.C,WangK.Q,et al.A Hierachical multiscale and multiangle system of for human face detection in a complex background using gravity-center template. Pattern Recongnition, 1999, 32(10): 1237-1248.
[26]艾海舟,肖习攀,徐光裕.人脸定位与检索.计算机学报[J],2003,26(7):874-880.
[27]李武军,钟翔平,李宁等.一种快速而精确的多人脸定位与定位算法[J].小型微型计算机系统,2005,26(9):1520-1524.
[28]冈萨雷斯.数字图像处理.第二版.电子工业出版社.224-232.
[29] Natravali, A.N.and Haskellm B.G.. Digital Pictures-Representation and Compression [M]. Plenum Press, New York and London, 1988
[30]陶霖密,彭振云,徐光枯.人体的肤色特征.软件学报,2001,7(12):1031-1041.
[31] Rafael C.Gonzalez Richad E.Woods.数字图像处理(第二版).阮秋琦,阮宇智.北京:电于工业出版社,2003.
[32] Hsu RL, Abdel-mottaleb M, A K Jain. Face detection in color images[R]. IEEE Trans. Pattern Analysis and Machine Intelligence, vol.24, no.5, May 2002,696-706.
[33] Novak C L, Shafer S A. Supervised Color constancy for maehine vision. In: Proeeedings of SPIE,HumanVision, Vision Processing, and DigitalDisPlay, New York,1991,1453:353-368.
[34] Aaron C Shumate, Hui Li, Color Blancing in Digital Cameras, http:Hise.stanford edu/class /phych.221/00/trek/.
[35]张宏林.visua1C++数字图像模式识别技术及工程实践.第一版.北京:人民邮电出版社,2003.
[36]李介谷.图像处理技术.上海:上海交通大学出版社,2006.
[37] Milan Sonka, Vaclav Hlavac, Roger Boyle. Image Processing. Analysis, and Machine Vision[M].2 nd ed. Beijing: Post and telcom press,2003.
[38] M.Hu, S.Worall, A.H.Sadka, A.A.M.kondoz. Face feature detection and modual design for 2-D scalable face modual-based video codeing, VIE, Guildford-London, UK July 2003.
[39] M.Hu, S.Worall, A.H.Sadka, A.A.M.kondoz. A fast and efficient chin detection method for 2-D scalable face modual design, Guildford-London, UK July 2003.
[40] Novak C L, Shafer S A. Supervised Color constancy for machine vision. In: Processings of SPLE, Human Vision, Vision Processing, and Digital Display, New York, 1991, 1453:353-368.
[41] R.E.Schapire, Y.Freund. A decision-theoretic generarion of on-line learning and an application to boosting. Joural of Computer and System Sciences, 1997, 55(91):119-139.
[42]涂承胜,刁力力,鲁明羽等.Boosting家族AdaBoost系列代表算法.计算机科学,2003,3(30):30-35.
[43]涂承胜,陆玉昌.Boosting理论基础.计算机科学,2004,10(31):11-14.
[44]张成功.基于AdoBoost算法的自动人脸定位与识别:(硕士学位论文).天津:天津师范大学,2008,6.
[2]梁路宏,艾海舟,徐光佑等.人脸定位研究综述.计算机学报,2002,5(25):449-459.
[3]周杰,卢春雨,张长水等.人脸自动识别方法综述.电子学报,2000,28(4):102-106.
[4]贾云得.机器视觉.北京:科学出版社,2000.
[5]江嘉明.基于自然图片的人脸定位识别系统的实现:(硕士学位论文).大连:大连理工大学,2008,6.
[6] Brunelli R,Poggiot T.Template matching method resarch.. New York: Academic Pr,1999.
[7] Lee CH,KimJS,Park KH. Automatic human face location in a complex background using motion and color information.Pattern Recognition,1996,29(11):1877-1899.
[8] DalY, NakanoY.Face texture model based on SGLD and application in face detection in a color scene. Pattern Recognition,1996,29(6):1007-1017.
[9] Moghaddam B,Pentland A. Probabilistic Visual Learning for Object Representation. IEEE Transon Pattern Analysis and Machine Intelligence,1997,19(7):696-710.
[10] Rowley HA, Baluja S, Kanade T. Rotation invariant neural network-based face detection, CM U-CS-97-201, 1997.
[11] Saber E,Tekalp AM.Frontal-view face detection and facial feature extraction using color shape and symmetry based cost functions[J]. Pattern Recognition Letters, 1998, 19(8): 669-680.
[12]邓刚,闫胜业,张洪明。人脸定位技术报告,2001.
[13]罗敏.基于肤色的人脸定位和面部特征定位技术研究:(硕士学位论文).江苏:江苏大学,2008,5.
[14]唐伟,陈兆乾,吴建鑫等.静态灰度图像中的人脸定位方法综述.计算机科学,2002,29(2):134-137.
[15] Yang G, Huang T.s.Human face detection in complex background.Pattern Recognition. 1994,27(1):53-63.
[16] Kotropulos C., Pitas I. Rule-based face detection in frontal viwes. In Proceedings of International Conference on Acoustics, Speech and Singal Processing, 1996, 4:312-317.
[17] DaiY., NakanoY, Extraction for facial images from complex background using color information and SGLD matrices.In Proceeding of the First International Workshop on Automatic Face and Gesture Recognition,1997,(4):2537-2540.
[18] Turk M., Pentland A.Eingenface for recognition .Journal of congnitive neuroscience, 1991, (1): 71-86.
[19] Pentland A, MoghaddamB, StarnerT.View_based and modular eigenspaces for face recongnition. Seattle, WA, 1994.21-23.
[20] Osuna E, Freund R, Girosi F.Training support vector machines: an application to face detection. Proceddings of Computer Vision and Pattern Recongnition Puerto Rico, 1997.130-136.
[21] Viola P, Jones M.Robust real time objects detection. International Journal of Computer Vision 2002, 57(2):13-7154.
[22] Hsu.R.L, Mohamed Abdel-Mottaleb M, Jain A.K. Face Detection in Color Images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24(5): 696-706.
[23] Yow K.C, Cipolla R.Feauter-based human face detection. Image and Vision Computing, 1997, 15(9): 713-735.
[24] YullieA, HallinanP, CohenD. Feature exaction from faces using deformable templates. Internation al Journal of Computing Vision, 1992, 8(2): 99-111.
[25] Miao J, YinB.C,WangK.Q,et al.A Hierachical multiscale and multiangle system of for human face detection in a complex background using gravity-center template. Pattern Recongnition, 1999, 32(10): 1237-1248.
[26]艾海舟,肖习攀,徐光裕.人脸定位与检索.计算机学报[J],2003,26(7):874-880.
[27]李武军,钟翔平,李宁等.一种快速而精确的多人脸定位与定位算法[J].小型微型计算机系统,2005,26(9):1520-1524.
[28]冈萨雷斯.数字图像处理.第二版.电子工业出版社.224-232.
[29] Natravali, A.N.and Haskellm B.G.. Digital Pictures-Representation and Compression [M]. Plenum Press, New York and London, 1988
[30]陶霖密,彭振云,徐光枯.人体的肤色特征.软件学报,2001,7(12):1031-1041.
[31] Rafael C.Gonzalez Richad E.Woods.数字图像处理(第二版).阮秋琦,阮宇智.北京:电于工业出版社,2003.
[32] Hsu RL, Abdel-mottaleb M, A K Jain. Face detection in color images[R]. IEEE Trans. Pattern Analysis and Machine Intelligence, vol.24, no.5, May 2002,696-706.
[33] Novak C L, Shafer S A. Supervised Color constancy for maehine vision. In: Proeeedings of SPIE,HumanVision, Vision Processing, and DigitalDisPlay, New York,1991,1453:353-368.
[34] Aaron C Shumate, Hui Li, Color Blancing in Digital Cameras, http:Hise.stanford edu/class /phych.221/00/trek/.
[35]张宏林.visua1C++数字图像模式识别技术及工程实践.第一版.北京:人民邮电出版社,2003.
[36]李介谷.图像处理技术.上海:上海交通大学出版社,2006.
[37] Milan Sonka, Vaclav Hlavac, Roger Boyle. Image Processing. Analysis, and Machine Vision[M].2 nd ed. Beijing: Post and telcom press,2003.
[38] M.Hu, S.Worall, A.H.Sadka, A.A.M.kondoz. Face feature detection and modual design for 2-D scalable face modual-based video codeing, VIE, Guildford-London, UK July 2003.
[39] M.Hu, S.Worall, A.H.Sadka, A.A.M.kondoz. A fast and efficient chin detection method for 2-D scalable face modual design, Guildford-London, UK July 2003.
[40] Novak C L, Shafer S A. Supervised Color constancy for machine vision. In: Processings of SPLE, Human Vision, Vision Processing, and Digital Display, New York, 1991, 1453:353-368.
[41] R.E.Schapire, Y.Freund. A decision-theoretic generarion of on-line learning and an application to boosting. Joural of Computer and System Sciences, 1997, 55(91):119-139.
[42]涂承胜,刁力力,鲁明羽等.Boosting家族AdaBoost系列代表算法.计算机科学,2003,3(30):30-35.
[43]涂承胜,陆玉昌.Boosting理论基础.计算机科学,2004,10(31):11-14.
[44]张成功.基于AdoBoost算法的自动人脸定位与识别:(硕士学位论文).天津:天津师范大学,2008,6.