嵌入式系统上基于近红外图像的人脸检测的研究
|
摘要
人脸检测的任务是对于一个输入图像,给出图像中是否存在人脸的判断,如果存在人脸,给出人脸的具体位置与范围。人脸检测是人脸识别技术的一个重要组成部分,随着时代的发展,已逐渐发展成一个重要的研究领域。人脸检测的难点包括:光照影响、年龄影响、装饰影响、姿态影响等。对于嵌入式设备,时间开销与内存开销直接决定了检测算法的可用性。
为了减少环境光照的影响,工业界与学术界都进行了大量的研究工作。相比算法的改进,更为有效的办法是利用主动近红外光进行人脸检测和人脸识别。
本文的主要工作是将人脸检测方法进行了实验比较,实现了一个包含多算法的Boosting训练器,提供了训练专用的SDK。为了减少环境光带来的影响,引入了主动近红外光的光照系统和专用的摄像头模组。在完成人脸区域定位的研究工作后,进行了人眼精确定位的研究,并与其他定位算法进行了比较。为了将检测算法应用于嵌入式系统,本文提出了很多的优化方案,对于程序执行的时间开销有很好的优化效果。
本文还提出了一种新型的基于数理形态学的近红外人脸检测算法。算法首先使用基于Haar特征和AdaBoost算法进行检测器的训练,在完成训练后利用分类器对目标图像寻找人脸候选区域,然后对候选区域进行归一化处理。利用人眼瞳孔在近红外光图像中会生成白色光斑的特点,使用基于数理形态学的Quoit滤波器精确定位眼睛。由于每个人瞳孔大小有差导,单一的Quoit滤波器无法适应这种情况。为了减小不同瞳孔大小带来的影响,使用了多尺度的Quoit滤波器以提高准确性。
在实验部分,本文对训练算法的性能进行了评估。实验表明,AdaBoost+LDA的训练算法具有最快的错误率收敛速度。同时,引入图像处理后,级联分类器的错误率也有很大的改进。多尺度Quoit滤波器与AdaBoost人眼定位方法相比,有更好的检测率和定位精度。总而言之,本文提出的人脸检测算法不仅准确性高而且速度快,达到了实时人脸检测的要求。
The task of face detection is to indicate whether there exist faces on specified image. If there are faces, the detection system should return the coordinate and range of faces. Face detection is an important part of face recognition technology. As the time goes by, it has been developed to an important course. The difficulties of face detection are light condition, age changing, accessory changing, attitude changing and etc. For embedded system, the cost of time and memory determine the availability of detection algorithm.
In order to reduce the affect cause by light condition, lots of researches have been done in both industry and academic. Compare with improvement of algorithm, the more effective way is to use active Near-Infrared light system to perform face detection and recognition.
This paper compares many face detection algorithms base on experience. Implements a Boosting trainer includes many algorithms, provides training SDK. To reduce the light condition affect, active Near-Infrared light system and camera are used. After the work of face area location, eye detection is also researched and compares the algorithm with other algorithm. To make the algorithm work on embedded system, this paper proposed lots of improved methods. It shows that it can improve the cost of time.
This paper also formulates a face detection algorithm base on morphology. Face detection base on Haar feature and AdaBoost algorithm is used for locating the face area. And then normalize the area into specify size. Using the property of high reflection rate under Near-Infrared light on pupil, Quoit filter base on morphology is used for eye detection. In order to dealing with difference size of pupil, a multi-scale filter is proposed for reducing both of false positive rate and false negative rate.
The experience shows that AdaBoost+LDA training method can lower the error rate more quickly than others. On the other hand, uses image processing on detection system can improve the detection rate. Multi-scale Quoit filter eye detection algorithm is more precise than AdaBoost eye detection. To conclude, the method proposed by this article is accuracy and fast. And it fits the requirement of real time face detection.
为了减少环境光照的影响,工业界与学术界都进行了大量的研究工作。相比算法的改进,更为有效的办法是利用主动近红外光进行人脸检测和人脸识别。
本文的主要工作是将人脸检测方法进行了实验比较,实现了一个包含多算法的Boosting训练器,提供了训练专用的SDK。为了减少环境光带来的影响,引入了主动近红外光的光照系统和专用的摄像头模组。在完成人脸区域定位的研究工作后,进行了人眼精确定位的研究,并与其他定位算法进行了比较。为了将检测算法应用于嵌入式系统,本文提出了很多的优化方案,对于程序执行的时间开销有很好的优化效果。
本文还提出了一种新型的基于数理形态学的近红外人脸检测算法。算法首先使用基于Haar特征和AdaBoost算法进行检测器的训练,在完成训练后利用分类器对目标图像寻找人脸候选区域,然后对候选区域进行归一化处理。利用人眼瞳孔在近红外光图像中会生成白色光斑的特点,使用基于数理形态学的Quoit滤波器精确定位眼睛。由于每个人瞳孔大小有差导,单一的Quoit滤波器无法适应这种情况。为了减小不同瞳孔大小带来的影响,使用了多尺度的Quoit滤波器以提高准确性。
在实验部分,本文对训练算法的性能进行了评估。实验表明,AdaBoost+LDA的训练算法具有最快的错误率收敛速度。同时,引入图像处理后,级联分类器的错误率也有很大的改进。多尺度Quoit滤波器与AdaBoost人眼定位方法相比,有更好的检测率和定位精度。总而言之,本文提出的人脸检测算法不仅准确性高而且速度快,达到了实时人脸检测的要求。
The task of face detection is to indicate whether there exist faces on specified image. If there are faces, the detection system should return the coordinate and range of faces. Face detection is an important part of face recognition technology. As the time goes by, it has been developed to an important course. The difficulties of face detection are light condition, age changing, accessory changing, attitude changing and etc. For embedded system, the cost of time and memory determine the availability of detection algorithm.
In order to reduce the affect cause by light condition, lots of researches have been done in both industry and academic. Compare with improvement of algorithm, the more effective way is to use active Near-Infrared light system to perform face detection and recognition.
This paper compares many face detection algorithms base on experience. Implements a Boosting trainer includes many algorithms, provides training SDK. To reduce the light condition affect, active Near-Infrared light system and camera are used. After the work of face area location, eye detection is also researched and compares the algorithm with other algorithm. To make the algorithm work on embedded system, this paper proposed lots of improved methods. It shows that it can improve the cost of time.
This paper also formulates a face detection algorithm base on morphology. Face detection base on Haar feature and AdaBoost algorithm is used for locating the face area. And then normalize the area into specify size. Using the property of high reflection rate under Near-Infrared light on pupil, Quoit filter base on morphology is used for eye detection. In order to dealing with difference size of pupil, a multi-scale filter is proposed for reducing both of false positive rate and false negative rate.
The experience shows that AdaBoost+LDA training method can lower the error rate more quickly than others. On the other hand, uses image processing on detection system can improve the detection rate. Multi-scale Quoit filter eye detection algorithm is more precise than AdaBoost eye detection. To conclude, the method proposed by this article is accuracy and fast. And it fits the requirement of real time face detection.
引文
[1] YANG MH, KRIEGMAN DJ, and AHUJA N. Detecting faces in image: A survey[J]. IEEE Transactions on Pattern analysis and Machine intelligence. 2002,24(1):34-58.
[2] ROWLEY H A, BALUJA S, KANADE T. Neural network-based face detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence.1998,20(1): 23-38.
[3] OSUNA E, FREUND R, and GIROSI F. Training support vector machines: an application to face detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Puerto Rico:IEEE, 1997:130-136.
[4] VIOLA P, JONES M. Robust real-time object detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Kauai:IEEE,2001.
[5] LI S Z, Zhang Z Q. FloatBoost learning and statistical face detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2004: 1112-1123.
[6] HUANG C, AI H, Li Y, et al. High-performance rotation invariant multiview face detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence.2007.29(4): 671-686.
[7] PHAM M T, CHAM T J. Fast training and selection of Haar features using statistics in boosting-based face detection[C]//IEEE 11~(th) International Conference on Computer Vision. [s.l.]:IEEE, 2007.
[8] WANG P, GREEN M B, JI Q, et al. Automatic eye detection and its validation[C]//IEEE International Conference on Computer Vision and Pattern Recognition. San Diego:IEEE, 2005:164-164.
[9] GUAN Y. Robust eye detection from facial image based on multi-cue facial information[C]//IEEE International Conference on Control and Automation.GuangZhou:IEEE, 2007:1775-1778
[10]SCHAPIRE R E, SINGER Y. Improved boosting algorithm using confidence-rated predictions. Proc. 11th Ann. Conf. Computational Learning Theory. 1998: 80-91.
[11] GONZALEZ R C, WOODS R E. Digital image processing second edition[M].Bejing Publishing House of Electronics Industry, 2002.
[12] VIOLA P and JONES M. Fast and Robust Classification using Asymmetric AdaBoost and a Detector Cascade[J]. Advances in Neural Information Processing Systems.2002.2:1311 - 1318.
[13]SCHAPIRE RE,FREUND Y,BARTLETT P,and LEE WS.Boosting the margin:a new explanation for the effectiveness of voting methods[J].The Annals of Statistics.1998.26(5):1651 - 1686
[14]WU J,BRUBAKER SC,MULLIN MD,etal.Fast Asymmetric Learning for Cascade Face Detection.IEEE Transactions on Pattern Analysis and Machine Intelligence.2008.30(3):369.
[15]LI SZ,JAIN AK,etal.Handbook of Face Recognition[M].[s.l.]:Springer-Verlag,2004.
[16]边肇棋,张学工等.模式识别(第二版)[M].北京:清华大学出版社,2000年1月第2版.
[17]FRIEDMAN J,HASTIE T and TIBSHIRANI R.Additive logistic regression:a statistical view of boosting[J].The Annals of Statistics.2000.28(2):337-407.
[18]FREUND Y and SCHAPIRE RE.A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting[J].Journal of Computer and System Sciences.1997.55(2):119-139.
[19]PHILLIPS PJ,SCRUGGS WT,O'TOOLE AJ,et al.FRVT 2006 and ICE 2006Large-Scale Results[R/OL].[s.l.]:NISTIR 7408,2007.http://www.frvt.org/FRVT2006/docs/FRVT2006an dICE2006LargeScaleReport.pdf
[20]RODHULLANDEMU,et al.Infrared[EB/OL].[s.l.]:WIKI,2009.http://en.wikipedi a.org/wiki/Infrared
[21]ZORROBOT,et a1.仿射变换[EB/OL].[s.l.]:WIKI CN,2009.http://zh.wikipedia.org/wiki/%E4%BB%BF%E5%B0%84%E5%8F%98%E6%8D%A2
[22]JANDBOT,et a1.双线性插值[EB/OL].[s.l.]:WIKI CN,2009.http://zh.wikipedia.org/wiki/%E5%8F%8C%E7%BA%BF%E6%80%A7%E6%8F%92%E5%80%BC
[23]YANGJEEP,et al.直方图均衡化[EB/OL].[s.1.]:WIKI CN,2009.http://zh.wiki pedia.org/wiki/%E7%9B%B4%E6%96%B9%E5%9B%BE%E5%9D%87%E8%A1%A1%E5%8C%96
[24]Yu SQ,et al.OpenCV 中文网站[EB/OL].[s.l.]:[s.n.]i,2009.http://www.opencv.org.cn/
[25]StarUML Org.StarUML[EB/OL].[s.l.]:[s.n.],2009.http://staruml.sourceforge.net/
[26]GONZALEZ RC,WOODS RE and EDDINS SL.Digital Image Processing Using MATLAB[M].Bejing:Publishing House of Electronics Industry.2005
[27]Xie SF.CAS-PEAL Face Database[EB/OL].Beijing:Chinese Academy of Sciences.2009.http://www.jdl.ac.cn/peal/index.html
[28]Tong wQ,et a1.上海红格电子科技有限公司.[s.l.]:[s.n.],2009.http://www.hongeltec.cn/
[29]王基帆,钱艺,童卫青.基于AdaBoost算法的人眼定位[J].现代计算机.2009.9:73-75.
[30]FARRUGIA N,MAMALET F,ROUX S,et al.Fast and Robust Face Detection on a Parallel Optimized Architecture Implemented on FPGA[J].IEEE Transactions on circuits and system for video technology.2009.19(4):597-602.
[31]THEOCHARIDES T,LINK G,VIJAYKRISHNAN N,et al.Embedded hardware face detection[C]//Proceedings of the 17th International Conference on VLSI Design.Mumbai,India:[s.n.],2004:133-138.
[32]王基帆,童卫青.基于数理形态学的近红外光图像实时人脸检测.华东师范大学学报自然科学版.2010.3
[2] ROWLEY H A, BALUJA S, KANADE T. Neural network-based face detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence.1998,20(1): 23-38.
[3] OSUNA E, FREUND R, and GIROSI F. Training support vector machines: an application to face detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Puerto Rico:IEEE, 1997:130-136.
[4] VIOLA P, JONES M. Robust real-time object detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Kauai:IEEE,2001.
[5] LI S Z, Zhang Z Q. FloatBoost learning and statistical face detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2004: 1112-1123.
[6] HUANG C, AI H, Li Y, et al. High-performance rotation invariant multiview face detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence.2007.29(4): 671-686.
[7] PHAM M T, CHAM T J. Fast training and selection of Haar features using statistics in boosting-based face detection[C]//IEEE 11~(th) International Conference on Computer Vision. [s.l.]:IEEE, 2007.
[8] WANG P, GREEN M B, JI Q, et al. Automatic eye detection and its validation[C]//IEEE International Conference on Computer Vision and Pattern Recognition. San Diego:IEEE, 2005:164-164.
[9] GUAN Y. Robust eye detection from facial image based on multi-cue facial information[C]//IEEE International Conference on Control and Automation.GuangZhou:IEEE, 2007:1775-1778
[10]SCHAPIRE R E, SINGER Y. Improved boosting algorithm using confidence-rated predictions. Proc. 11th Ann. Conf. Computational Learning Theory. 1998: 80-91.
[11] GONZALEZ R C, WOODS R E. Digital image processing second edition[M].Bejing Publishing House of Electronics Industry, 2002.
[12] VIOLA P and JONES M. Fast and Robust Classification using Asymmetric AdaBoost and a Detector Cascade[J]. Advances in Neural Information Processing Systems.2002.2:1311 - 1318.
[13]SCHAPIRE RE,FREUND Y,BARTLETT P,and LEE WS.Boosting the margin:a new explanation for the effectiveness of voting methods[J].The Annals of Statistics.1998.26(5):1651 - 1686
[14]WU J,BRUBAKER SC,MULLIN MD,etal.Fast Asymmetric Learning for Cascade Face Detection.IEEE Transactions on Pattern Analysis and Machine Intelligence.2008.30(3):369.
[15]LI SZ,JAIN AK,etal.Handbook of Face Recognition[M].[s.l.]:Springer-Verlag,2004.
[16]边肇棋,张学工等.模式识别(第二版)[M].北京:清华大学出版社,2000年1月第2版.
[17]FRIEDMAN J,HASTIE T and TIBSHIRANI R.Additive logistic regression:a statistical view of boosting[J].The Annals of Statistics.2000.28(2):337-407.
[18]FREUND Y and SCHAPIRE RE.A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting[J].Journal of Computer and System Sciences.1997.55(2):119-139.
[19]PHILLIPS PJ,SCRUGGS WT,O'TOOLE AJ,et al.FRVT 2006 and ICE 2006Large-Scale Results[R/OL].[s.l.]:NISTIR 7408,2007.http://www.frvt.org/FRVT2006/docs/FRVT2006an dICE2006LargeScaleReport.pdf
[20]RODHULLANDEMU,et al.Infrared[EB/OL].[s.l.]:WIKI,2009.http://en.wikipedi a.org/wiki/Infrared
[21]ZORROBOT,et a1.仿射变换[EB/OL].[s.l.]:WIKI CN,2009.http://zh.wikipedia.org/wiki/%E4%BB%BF%E5%B0%84%E5%8F%98%E6%8D%A2
[22]JANDBOT,et a1.双线性插值[EB/OL].[s.l.]:WIKI CN,2009.http://zh.wikipedia.org/wiki/%E5%8F%8C%E7%BA%BF%E6%80%A7%E6%8F%92%E5%80%BC
[23]YANGJEEP,et al.直方图均衡化[EB/OL].[s.1.]:WIKI CN,2009.http://zh.wiki pedia.org/wiki/%E7%9B%B4%E6%96%B9%E5%9B%BE%E5%9D%87%E8%A1%A1%E5%8C%96
[24]Yu SQ,et al.OpenCV 中文网站[EB/OL].[s.l.]:[s.n.]i,2009.http://www.opencv.org.cn/
[25]StarUML Org.StarUML[EB/OL].[s.l.]:[s.n.],2009.http://staruml.sourceforge.net/
[26]GONZALEZ RC,WOODS RE and EDDINS SL.Digital Image Processing Using MATLAB[M].Bejing:Publishing House of Electronics Industry.2005
[27]Xie SF.CAS-PEAL Face Database[EB/OL].Beijing:Chinese Academy of Sciences.2009.http://www.jdl.ac.cn/peal/index.html
[28]Tong wQ,et a1.上海红格电子科技有限公司.[s.l.]:[s.n.],2009.http://www.hongeltec.cn/
[29]王基帆,钱艺,童卫青.基于AdaBoost算法的人眼定位[J].现代计算机.2009.9:73-75.
[30]FARRUGIA N,MAMALET F,ROUX S,et al.Fast and Robust Face Detection on a Parallel Optimized Architecture Implemented on FPGA[J].IEEE Transactions on circuits and system for video technology.2009.19(4):597-602.
[31]THEOCHARIDES T,LINK G,VIJAYKRISHNAN N,et al.Embedded hardware face detection[C]//Proceedings of the 17th International Conference on VLSI Design.Mumbai,India:[s.n.],2004:133-138.
[32]王基帆,童卫青.基于数理形态学的近红外光图像实时人脸检测.华东师范大学学报自然科学版.2010.3