两层级联卷积神经网络的人脸检测
|
摘要
目的传统人脸检测方法因人脸多姿态变化和人脸面部特征不完整等问题,导致检测效果不佳。为解决上述问题,提出一种两层级联卷积神经网络(TC_CNN)人脸检测方法。方法首先,构建两层卷积神经网络模型,利用前端卷积神经网络模型对人脸图像进行特征粗略提取,再利用最大值池化方法对粗提取得到的人脸特征进行降维操作,输出多个疑似人脸窗口;其次,将前端粗提取得到的人脸窗口作为后端卷积神经网络模型的输入进行特征精细提取,并通过池化操作得到新的特征图;最后,通过全连接层判别输出最佳检测窗口,完成人脸检测全过程。结果实验选取FDDB人脸检测数据集中包含人脸多姿态变化以及人脸面部特征信息不完整等情况的图像进行测试,TC_CNN方法人脸检测率达到96. 39%,误检率低至3. 78%,相比当前流行方法在保证算法效率的同时检测率均有提高。结论两层级联卷积神经网络人脸检测方法能够在人脸多姿态变化和面部特征信息不完整等情况下实现精准检测,保证较高的检测率,有效降低误检率,方法具有较好的鲁棒性和泛化能力。
Objective As an important part of face recognition,face detection has attracted considerable attention in comput-er vision and has been widely investigated. Face detection determines the location and size of human faces in an image.Traditional face detection methods are limited by face multi-pose changes and incomplete facial features,which lead to theirpoor detection effect. Modern face detectors can easily detect near-frontal faces. Recent research in this area has focused onthe uncontrolled face detection problem,where a number of factors,such as multi-pose changes and incomplete facial fea-tures,can lead to large visual variations in face appearance and can severely degrade the robustness of the face detector. Aconvolutional neural network can automatically select facial features,rapidly delete a large number of non-face backgroundinformation,and can achieve good face detection results. However,a single convolutional neural network should possessthree functions,namely,facial feature extraction,reduction of feature dimensions to decrease the computational complexi-ty,and feature classification,which result in complex network structure,limited detection speed,and overfitting of the net-work. To solve these problems,this study presents a face detection method of two-layer cascaded convolutional neural net-work( TC_ CNN). Method First,a two-layer convolutional neural network model is constructed. The first convolutionalneural network model is used to extract the features of the face image,and a max pooling method is adopted to reduce thedimension of those features in which multiple suspected face windows are outputted. Second,the face windows are used asthe inputs of the second convolutional neural network model for fine feature extraction,and a new feature map is obtained bypool operation. Finally,the best detection window is outputted through full connection layer discrimination. The face issuccessfully detected and the face window is returned when the result of discriminant classification is a face; otherwise,thenon-face window is deleted. An optimal face detection window can be selected through non-maximum suppression,the sizeand position of the face in the input image are returned based on the location information of the optimal face detection win-dow,and the entire process of face detection is completed. In the training of TC_ CNN,we use 10 000 images with near-frontal faces,face multi-pose changes,and incomplete facial features from the labeled faces in the Wild dataset as positivetraining samples and 1 000 images as negative training samples. In the testing of the TC_ CNN model,we utilize an authori-tative dataset FDDB to evaluate,measure,and determine the validity of the model based on four indexes,namely,detec-tion rate,false detection rate,missing detection rate,and detection time. The TC_ CNN model is compared with excellentface detection algorithms,such as Ada Boost,fast LBP,NPD + Ada Boost,and SPP + CNN methods. Result Images withface multi-pose changes and incomplete face feature information in the FDDB face detection dataset are selected for the test.Results show that the face detection rate by TC_ CNN method is up to 96. 39%,false detection rate is as low as 3. 78%,and detection time is 0. 451 s. For the detection rate,the TC_ CNN method is 7. 63% higher than the traditional Ada Boostmethod based on cascade idea,3. 57% higher than the fast LBP method,0. 50% higher than the NPD + Ada Boost method,and 6. 04% higher than the SPP + CNN method. For the false detection rate,the TC_ CNN method is 2. 44% lower thanthe Ada Boost method,4. 47% lower than the fast LBP method,0. 59% lower than the NPD + Ada Boost method,and5. 09% lower than the SPP + CNN method. For the detection time,the TC_ CNN method' s detection efficiency is remark-ably higher than the SPP + CNN method and slightly higher than the Ada Boost,fast LBP,and NPD + Ada Boost methods. Incomparison with the current methods,the detection rate is increased while ensuring the efficiency of the algorithm. To verifythe robustness of the TC_ CNN model under the conditions of face multi-pose changes and incomplete facial features,repre-sentative images of two special cases are selected from the FDDB dataset in conducting four groups of comparative experi-ments under the multi-pose changes of a single face image,multi-pose changes of a multi face image,incomplete facial fea-tures of a single face image,and incomplete facial features of a multi face image. Experimental results show that theTC_ CNN model shows good effectiveness and robustness compared with the four excellent algorithms or four groups of cont-rastive experiments under different interference conditions. Conclusion The TC_ CNN model for face detection can achieveaccurate detection under face multi-pose changes and incomplete facial feature information. This model can obtain a highdetection rate and effectively reduce false detection rate. The method has good robustness and generalization capability. TheTC_ CNN method overcomes the limitations of the excellent Ada Boost cascade concept on the face detection method( suchas cascading two convolutional neural networks; effectively avoiding the complex network structure caused by the three func-tions of extraction,reduction,and classification of features simultaneously),which easily cause overfitting and other con-tradictions. However,the selection of the number and parameter of the cascaded convolutional neural network is difficult forthe improvement of the model performance and detection effect. In future research,we will determine the number andparameter of cascaded convolution neural network to optimize the model and will attempt to detect the size and position ofthe face accurately.
Objective As an important part of face recognition,face detection has attracted considerable attention in comput-er vision and has been widely investigated. Face detection determines the location and size of human faces in an image.Traditional face detection methods are limited by face multi-pose changes and incomplete facial features,which lead to theirpoor detection effect. Modern face detectors can easily detect near-frontal faces. Recent research in this area has focused onthe uncontrolled face detection problem,where a number of factors,such as multi-pose changes and incomplete facial fea-tures,can lead to large visual variations in face appearance and can severely degrade the robustness of the face detector. Aconvolutional neural network can automatically select facial features,rapidly delete a large number of non-face backgroundinformation,and can achieve good face detection results. However,a single convolutional neural network should possessthree functions,namely,facial feature extraction,reduction of feature dimensions to decrease the computational complexi-ty,and feature classification,which result in complex network structure,limited detection speed,and overfitting of the net-work. To solve these problems,this study presents a face detection method of two-layer cascaded convolutional neural net-work( TC_ CNN). Method First,a two-layer convolutional neural network model is constructed. The first convolutionalneural network model is used to extract the features of the face image,and a max pooling method is adopted to reduce thedimension of those features in which multiple suspected face windows are outputted. Second,the face windows are used asthe inputs of the second convolutional neural network model for fine feature extraction,and a new feature map is obtained bypool operation. Finally,the best detection window is outputted through full connection layer discrimination. The face issuccessfully detected and the face window is returned when the result of discriminant classification is a face; otherwise,thenon-face window is deleted. An optimal face detection window can be selected through non-maximum suppression,the sizeand position of the face in the input image are returned based on the location information of the optimal face detection win-dow,and the entire process of face detection is completed. In the training of TC_ CNN,we use 10 000 images with near-frontal faces,face multi-pose changes,and incomplete facial features from the labeled faces in the Wild dataset as positivetraining samples and 1 000 images as negative training samples. In the testing of the TC_ CNN model,we utilize an authori-tative dataset FDDB to evaluate,measure,and determine the validity of the model based on four indexes,namely,detec-tion rate,false detection rate,missing detection rate,and detection time. The TC_ CNN model is compared with excellentface detection algorithms,such as Ada Boost,fast LBP,NPD + Ada Boost,and SPP + CNN methods. Result Images withface multi-pose changes and incomplete face feature information in the FDDB face detection dataset are selected for the test.Results show that the face detection rate by TC_ CNN method is up to 96. 39%,false detection rate is as low as 3. 78%,and detection time is 0. 451 s. For the detection rate,the TC_ CNN method is 7. 63% higher than the traditional Ada Boostmethod based on cascade idea,3. 57% higher than the fast LBP method,0. 50% higher than the NPD + Ada Boost method,and 6. 04% higher than the SPP + CNN method. For the false detection rate,the TC_ CNN method is 2. 44% lower thanthe Ada Boost method,4. 47% lower than the fast LBP method,0. 59% lower than the NPD + Ada Boost method,and5. 09% lower than the SPP + CNN method. For the detection time,the TC_ CNN method' s detection efficiency is remark-ably higher than the SPP + CNN method and slightly higher than the Ada Boost,fast LBP,and NPD + Ada Boost methods. Incomparison with the current methods,the detection rate is increased while ensuring the efficiency of the algorithm. To verifythe robustness of the TC_ CNN model under the conditions of face multi-pose changes and incomplete facial features,repre-sentative images of two special cases are selected from the FDDB dataset in conducting four groups of comparative experi-ments under the multi-pose changes of a single face image,multi-pose changes of a multi face image,incomplete facial fea-tures of a single face image,and incomplete facial features of a multi face image. Experimental results show that theTC_ CNN model shows good effectiveness and robustness compared with the four excellent algorithms or four groups of cont-rastive experiments under different interference conditions. Conclusion The TC_ CNN model for face detection can achieveaccurate detection under face multi-pose changes and incomplete facial feature information. This model can obtain a highdetection rate and effectively reduce false detection rate. The method has good robustness and generalization capability. TheTC_ CNN method overcomes the limitations of the excellent Ada Boost cascade concept on the face detection method( suchas cascading two convolutional neural networks; effectively avoiding the complex network structure caused by the three func-tions of extraction,reduction,and classification of features simultaneously),which easily cause overfitting and other con-tradictions. However,the selection of the number and parameter of the cascaded convolutional neural network is difficult forthe improvement of the model performance and detection effect. In future research,we will determine the number andparameter of cascaded convolution neural network to optimize the model and will attempt to detect the size and position ofthe face accurately.
引文
[1]Hao Z K,Liu Y,Qin H W,et al.Scale-aware face detection[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition.Honolulu,HI:IEEE,2017:1063-6919.[DOI:10.1109/CVPR.2017.207]
[2]Campos N C S,Monteiro H A,Brito A V,et al.A framework for design and validation of face detection systems[C]//Proceedings of 2017 CHILEAN Conference on Electrical,Electronics Engineering,Information and Communication Technologies.Pucon,Chile:IEEE,2017:1-7.[DOI:10.1109/CHILECON.2017.8229685]
[3]Mukherjee S,Saha S,Lahiri S,et al.Convolutional Neural Network based face detection[C]//Proceedings of 2017 1st International Conference on Electronics,Materials Engineering and Nano-Technology.Kolkata,India:IEEE,2017:1-5.[DOI:10.1109/IEMENTECH.2017.8076987]
[4]Zhang S,Cai Y,Xie M.Face detection based on local region sparse coding[J].Journal of Software,2013,24(11):2747-2757.[张抒,蔡勇,解梅.基于局部区域稀疏编码的人脸检测[J].软件学报,2013,24(11):2747-2757.][DOI:10.3724/SP.J.1001.2013.04484]
[5]Song R.The research and implementation of human face detection and tracking based on complex background[D].Hefei:Anhui University,2014.[宋茹.基于复杂场景的人脸检测与人脸跟踪的研究与实现[D].合肥:安徽大学,2014.]
[6]Bhandiwad V,Tekwani B.Face recognition and detection using neural networks[C]//Proceedings of 2017 International Conference on Trends in Electronics and Informatics.Tirunelveli,India:IEEE,2017:879-882.[DOI:10.1109/ICOEI.2017.8300832]
[7]Mohan K,Chandrasekhar P,Jilani S A K.A combined HOG-LPQ with Fuz-SVM classifier for Object face Liveness Detection[C]//Proceedings of 2017 International Conference on I-SMAC.Palladam,India:IEEE,2017:531-537.[DOI:10.1109/I-SMAC.2017.8058406]
[8]Yu M,Yun L J,Chen Z Q,et al.Research on video face detection based on Ada Boost algorithm training classifier[C]//Proceedings of 2017 First International Conference on Electronics Instrumentation&Information Systems.Harbin,China:IEEE,2017:1-6.[DOI:10.1109/EIIS.2017.8298627]
[9]Liu M Y.Face detection system based on static images[D].Chengdu:University of Electronic Science and Technology of China,2014.[刘明跃.基于静态图像的人脸检测[D].成都:电子科技大学,2014.]
[10]Malisiewicz T,Gupta A,Efros A A.Ensemble of exemplarSVMs for object detection and beyond[C]//Proceedings of 2011IEEE International Conference on Computer Vision.Barcelona,Spain:IEEE,2011:89-96.[DOI:10.1109/ICCV.2011.6126229]
[11]Wang Y Q.An analysis of the Viola-Jones face detection algorithm[J].Image Processing On Line,2014,4:128-148.[DOI:10.5201/ipol.2014.104]
[12]Tao Q Q.Face detection based on convolutional neural network and improved support vector machine[D].Hefei:Hefei University of Technology,2016.[陶勤勤.基于卷积神经网络和改进支持向量机的人脸检测[D].合肥:合肥工业大学,2016.]
[13]Li H X,Lin Z,Shen X H,et al.A convolutional neural network cascade for face detection[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition.Boston,MA:IEEE,2015:5325-5334.[DOI:10.1109/CVPR.2015.7299170]
[14]Qin H W,Yan J J,Li X,et al.Joint training of cascaded CNNfor face detection[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas,NV,USA:IEEE,2016:3456-3465.[DOI:10.1109/CVPR.2016.376]
[15]Egorov A D,Divitskii D U,Dolgih A A,et al.Some cases of optimization face detection methodes on image(Using the ViolaJones method as an example)[C]//Proceedings of 2018 IEEEConference of Russian Young Researchers in Electrical and Electronic Engineering.Moscow,Russia:IEEE,2018:1075-1078.[DOI:10.1109/EIConRus.2018.8317276]
[16]Krizhevsky A,Sutskever I,Hinton G E.Image Net classification with deep convolutional neural networks[C]//Proceedings of the25th International Conference on Neural Information Processing Systems.Lake Tahoe,Nevada:Curran Associates Inc.,2012:1097-1105.
[17]Ye X Y,Chen X T,Chen H H,et al.Cascaded probability state-restricted Boltzmann machine for face detection[J].Journal of Image and Graphics,2016,21(7):875-885.[叶学义,陈雪婷,陈华华,等.级联型P-RBM神经网络的人脸检测[J].中国图象图形学报,2016,21(7):875-885.][DOI:10.11834/jig.20160705]
[18]Tang Y T,Liang R,Liu C.Face detection system design based on the SOPC[J].Electronic Design Engineering,2014,(21):165-168.[汤欲涛,梁锐,刘畅.基于SOPC的人脸检测系统的设计[J].电子设计工程,2014,(21):165-168.]
[19]Bilaniuk O,Fazl-Ersi E,Laganière R,et al.Fast LBP face detection on low-power SIMD architectures[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition Workshops.Columbus,OH:IEEE,2014:616-622.[DOI:10.1109/CVPRW.2014.96]
[20]Micheal A A,Geetha P.Multi-view face detection using Normalized Pixel Difference feature[C]//Proceedings of 2017 International Conference on Communication and Signal Processing.Chennai,India:IEEE,2017:988-992.[DOI:10.1109/ICCSP.2017.8286520]
[21]Zheng C H,Liu B,Zhou Y.Face detection algorithm based on scale-independent cascade convolution neural network[J].Application Research of Computers,2019,36(3).[郑成浩,刘兵,周勇.尺度无关的级联卷积神经网络人脸检测算法[J].计算机应用研究,2019,36(3).]
[22]Lu H T,Zhang Q C.Applications of deep convolutional neural network in computer vision[J]//Journal of Data Acquisition and Processing,2016,31(1):1-17.[卢宏涛,张秦川.深度卷积神经网络在计算机视觉中的应用研究综述[J].数据采集与处理,2016,31(1):1-17.][DOI:10.16337/j.1004-9037.2016.01.001]
[2]Campos N C S,Monteiro H A,Brito A V,et al.A framework for design and validation of face detection systems[C]//Proceedings of 2017 CHILEAN Conference on Electrical,Electronics Engineering,Information and Communication Technologies.Pucon,Chile:IEEE,2017:1-7.[DOI:10.1109/CHILECON.2017.8229685]
[3]Mukherjee S,Saha S,Lahiri S,et al.Convolutional Neural Network based face detection[C]//Proceedings of 2017 1st International Conference on Electronics,Materials Engineering and Nano-Technology.Kolkata,India:IEEE,2017:1-5.[DOI:10.1109/IEMENTECH.2017.8076987]
[4]Zhang S,Cai Y,Xie M.Face detection based on local region sparse coding[J].Journal of Software,2013,24(11):2747-2757.[张抒,蔡勇,解梅.基于局部区域稀疏编码的人脸检测[J].软件学报,2013,24(11):2747-2757.][DOI:10.3724/SP.J.1001.2013.04484]
[5]Song R.The research and implementation of human face detection and tracking based on complex background[D].Hefei:Anhui University,2014.[宋茹.基于复杂场景的人脸检测与人脸跟踪的研究与实现[D].合肥:安徽大学,2014.]
[6]Bhandiwad V,Tekwani B.Face recognition and detection using neural networks[C]//Proceedings of 2017 International Conference on Trends in Electronics and Informatics.Tirunelveli,India:IEEE,2017:879-882.[DOI:10.1109/ICOEI.2017.8300832]
[7]Mohan K,Chandrasekhar P,Jilani S A K.A combined HOG-LPQ with Fuz-SVM classifier for Object face Liveness Detection[C]//Proceedings of 2017 International Conference on I-SMAC.Palladam,India:IEEE,2017:531-537.[DOI:10.1109/I-SMAC.2017.8058406]
[8]Yu M,Yun L J,Chen Z Q,et al.Research on video face detection based on Ada Boost algorithm training classifier[C]//Proceedings of 2017 First International Conference on Electronics Instrumentation&Information Systems.Harbin,China:IEEE,2017:1-6.[DOI:10.1109/EIIS.2017.8298627]
[9]Liu M Y.Face detection system based on static images[D].Chengdu:University of Electronic Science and Technology of China,2014.[刘明跃.基于静态图像的人脸检测[D].成都:电子科技大学,2014.]
[10]Malisiewicz T,Gupta A,Efros A A.Ensemble of exemplarSVMs for object detection and beyond[C]//Proceedings of 2011IEEE International Conference on Computer Vision.Barcelona,Spain:IEEE,2011:89-96.[DOI:10.1109/ICCV.2011.6126229]
[11]Wang Y Q.An analysis of the Viola-Jones face detection algorithm[J].Image Processing On Line,2014,4:128-148.[DOI:10.5201/ipol.2014.104]
[12]Tao Q Q.Face detection based on convolutional neural network and improved support vector machine[D].Hefei:Hefei University of Technology,2016.[陶勤勤.基于卷积神经网络和改进支持向量机的人脸检测[D].合肥:合肥工业大学,2016.]
[13]Li H X,Lin Z,Shen X H,et al.A convolutional neural network cascade for face detection[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition.Boston,MA:IEEE,2015:5325-5334.[DOI:10.1109/CVPR.2015.7299170]
[14]Qin H W,Yan J J,Li X,et al.Joint training of cascaded CNNfor face detection[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas,NV,USA:IEEE,2016:3456-3465.[DOI:10.1109/CVPR.2016.376]
[15]Egorov A D,Divitskii D U,Dolgih A A,et al.Some cases of optimization face detection methodes on image(Using the ViolaJones method as an example)[C]//Proceedings of 2018 IEEEConference of Russian Young Researchers in Electrical and Electronic Engineering.Moscow,Russia:IEEE,2018:1075-1078.[DOI:10.1109/EIConRus.2018.8317276]
[16]Krizhevsky A,Sutskever I,Hinton G E.Image Net classification with deep convolutional neural networks[C]//Proceedings of the25th International Conference on Neural Information Processing Systems.Lake Tahoe,Nevada:Curran Associates Inc.,2012:1097-1105.
[17]Ye X Y,Chen X T,Chen H H,et al.Cascaded probability state-restricted Boltzmann machine for face detection[J].Journal of Image and Graphics,2016,21(7):875-885.[叶学义,陈雪婷,陈华华,等.级联型P-RBM神经网络的人脸检测[J].中国图象图形学报,2016,21(7):875-885.][DOI:10.11834/jig.20160705]
[18]Tang Y T,Liang R,Liu C.Face detection system design based on the SOPC[J].Electronic Design Engineering,2014,(21):165-168.[汤欲涛,梁锐,刘畅.基于SOPC的人脸检测系统的设计[J].电子设计工程,2014,(21):165-168.]
[19]Bilaniuk O,Fazl-Ersi E,Laganière R,et al.Fast LBP face detection on low-power SIMD architectures[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition Workshops.Columbus,OH:IEEE,2014:616-622.[DOI:10.1109/CVPRW.2014.96]
[20]Micheal A A,Geetha P.Multi-view face detection using Normalized Pixel Difference feature[C]//Proceedings of 2017 International Conference on Communication and Signal Processing.Chennai,India:IEEE,2017:988-992.[DOI:10.1109/ICCSP.2017.8286520]
[21]Zheng C H,Liu B,Zhou Y.Face detection algorithm based on scale-independent cascade convolution neural network[J].Application Research of Computers,2019,36(3).[郑成浩,刘兵,周勇.尺度无关的级联卷积神经网络人脸检测算法[J].计算机应用研究,2019,36(3).]
[22]Lu H T,Zhang Q C.Applications of deep convolutional neural network in computer vision[J]//Journal of Data Acquisition and Processing,2016,31(1):1-17.[卢宏涛,张秦川.深度卷积神经网络在计算机视觉中的应用研究综述[J].数据采集与处理,2016,31(1):1-17.][DOI:10.16337/j.1004-9037.2016.01.001]
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |