基于流形结构的人脸民族特征研究

link管理

链接快照平台

输入网页链接，自动生成快照
标签化管理网页链接

Corresponding author: WANG Cun-Rui Associate professor at Dalian Minzu University. His research interest covers data mining and pattern recognition. Corresponding author of this paper

人脸民族特征选取与分析是人脸识别与人类学重要研究方向之一.本文建立了中国三个民族人脸数据库，通过流形结构来研究和分析人脸的民族特征.首先，在体质人类学定义的人脸几何特征指标进行流形分析，未形成按语义分布的子流形.因此本文将人脸特征扩至全部组合的长度、角度和比例特征进行分析，利用mRMR算法对2926个长度特征、21万余个角度特征、427万个比例特征中冗余特征进行筛选，加上人类学指标及混合筛选的数据集共形成5个数据集.利用LPP、Isomap、LE、PCA和LDA等流形方法分析5数据集，其中的4个数据集都形成了民族语义的子流形分布.为验证筛选特征指标的有效性，本文利用分类算法J48、SVM、RBF network、Naive Bayes、Bayes network在Weka平台对数据集以族群语义作为类别进行交叉验证实验，实验结果表明混合特征的人脸数据集族群分类平均准确率最高，且比例特征分类指标优于其他特征数据集.本文通过大量实验揭示了民族人脸数据可在子空间内形成按民族语义分布的子流形结构.中国三个民族人脸特征在低维空间存在不同民族语义的子流形，通过流形分析和特征筛选构建的人脸测量指标不仅可为人脸族群分析提供方法，同时也将丰富和补充体质人类学的相关研究工作. 人脸民族特征 / 生物特征识别 / 人脸识别 / Abstract: Facial ethnic feature selection and analysis is one of the most significant research focuses in face recognition and anthropology. In this paper, we build a Chinese ethnic face database including three ethnic groups. Manifold learning is used to analyze facial ethnic features. Firstly, we conduct manifold analysis on the basis of facial geometric indicators proposed by anthropologist, which, however, does not formulate sub-manifold distributed by semantics. Therefore, we intend to expand the scope of facial features by calculating the complete distances, angles and indexes with landmarks. Then, we adopt mRMR to filter 2926 distance indicators, more than 219450 angle indicators and more than 4279275 index indicators. Finally, we can obtain 5 datasets with features of distance, angle, index, anthropology and mixing. Several popular manifold learning methods including LPP, ISOMAP, LE, PCA and LDA are utilized to study the above mentioned datasets, and we get the distinguishable manifold structure of facial ethic feature and clusters in 4 of the 5 datasets. To evaluate the validity of filtered features, we make use of classification algorithms including J48, SVM, RBF network, Naive Bayes, and Bayes network implemented in Weka for cross validation experiments by ethnic semantics. Experimental results indicate that the average of classification accuracy on the dataset with mixing features is higher than that of other datasets, and that the index is more salient than other geometric features. Moreover, by full experimental investigation, we find that ethnic facial data can generate sub-manifold structure distributed by semantics. Facial features of three Chinese ethnic groups exhibit different ethnic semantic sub-manifolds in the low-dimensional space. Facial measurement indicators obtained by manifold analysis and feature selection not only provide a method for facial ethnic groups analysis, but also enrich and improve the related research work in anthropology. Key words: Facial ethnic features / biometrics recognition / face recognition / manifold learning (眼裂高度) / (眉眼距离)0.3291(眼裂高度) / (鼻翼与眉毛距离)0.362(鼻翼与眉毛距离)/ (嘴部与眉尖)0.312(鼻翼与眼内角点距离) / (额头高度)0.35(眼裂高度) / (鼻翼与眉毛距离)0.3022(鼻翼长度) / (眉眼距离)0.302(眉眼距离) / (眉毛与鼻翼距离)0.301(鼻翼长度) / (眉毛与嘴部距离)0.302(眼裂高度) / (鼻翼与眉毛距离)0.303(鼻翼与眼内角点距离) / (额头高度)0.294(鼻翼距离) / (嘴巴与眼外角点距离)0.297(眉间距) / (鼻翼与眼内角距离)0.297(眼裂高度) / (鼻翼与眼内角点距离)0.2744(眉毛与上唇距离) / (眉毛与下唇距离)0.283(鼻翼长度) / (眼睛与下颌距离)0.281 1I(49, 57)/(22, 7)0.66927I(39, 43)/(7, 22)0.2992I(35, 47)/(23, 51)0.36228I(49, 69)/(34, 72)0.2963I(37, 51)/(16, 24)0.3529I(22, 73)/(21, 64)0.2984I(39, 43)/(22, 36)0.32930I(49, 52)/(15, 7)0.2965I(50, 71)/(33, 60)0.3331I(35, 47)/(28, 51)0.2986I(49, 52)/(5, 17)0.31232I(25, 50)/(21, 27)0.2927I(22, 76)/(21, 54)0.31233I(37, 51)/(14, 19)0.2948I(51, 59)/(22, 45)0.30234A∠(21, 55, 26)0.2899I(31, 35)/(37, 51)0.30535I(39, 43)/(28, 51)0.28710A∠(51, 59, 27)0.31136I(49, 52)/(22, 38)0.28911I(39, 43)/(20, 58)0.30237I(49, 76)/(35, 72)0.28912I(37, 59)/(14, 22)0.30238I(50, 52)/(22, 60)0.28613I(17, 36)/(23, 50)0.30239I(35, 47)/(23, 50)0.28714I∠(31, 22, 33)0.29740I(49, 52)/(7, 35)0.28715I(49, 52)/(60, 74)0.30441I(22, 53)/(21, 50)0.28416I(50, 55)/(17, 55)0.30142I(50, 70)/(33, 60)0.28517I(18, 21)/(33, 49)0.30243A∠(17, 49, 21)0.28518I(35, 60)/(21, 54)0.30544I(37, 51)/(16, 24)0.28519I(39, 43)/(23, 51)0.30345I(37, 51)/(16, 24)0.28220I(37, 51)/(18, 25)0.30146A∠(51, 25, 59)0.28321I(22, 73)/(21, 76)0.34747A∠(35, 29, 49)0.28422I(49, 52)/(24, 66)0.30348I(49, 57)/(22, 43)0.28223I(49, 57)/(14, 22)0.30249I(39, 43)/(19, 49)0.28224I(50, 57)/(29, 61)0.29650A∠(21, 49, 25)0.28125A∠(21, 36, 22)0.29951I(31, 35)/(24, 51)0.27926A∠(22, 60, 50)0.298 注: I代表长度, A代表角度 139 ~ 436眼裂12216眉249 ~ 526鼻翼长度24916鼻337 ~ 514鼻眼距离35114鼻435 ~ 473眼裂42111眉549 ~ 573鼻翼宽度55010鼻622 ~ 732眉嘴距离6359眼731 ~ 352眼裂7377眼814 ~ 222额头高度18437眼916 ~ 242额头高度29527鼻1021 ~ 542眉鼻距离110396眼1123 ~ 502眉鼻距离211245眉1223 ~ 512眉鼻距离312574鼻13234眉14313眼15463眼16143额头17163额头18732嘴19542鼻 AM0.7530.1230.7530.7530.7530.814BM0.8330.0830.8340.8330.8330.879CM 0.92 0.04 0.921 0.921 0.921 0.935 DM0.900.050.9020.90.90 0.935 EM 0.96 0.02 0.96 0.96 0.96 0.975 AF0.7270.1370.7250.7270.7240.775BF0.7730.1130.7760.7730.7730.863CF 0.813 0.093 0.814 0.813 0.812 0.853 DF0.7670.1170.7650.7670.7640.844EF 0.813 0.093 0.818 0.813 0.814 0.888 AM0.820.090.8210.820.820.927BM0.900.050.9030.900.9010.96CM0.960.020.960.960.96 0.993 DM 0.967 0.017 0.968 0.967 0.967 0.992EM 0.973 0.013 0.974 0.973 0.973 0.999 AF0.7730.1130.7790.7730.7720.882BF0.7530.1230.7550.7530.7500.902CF 0.893 0.053 0.894 0.893 0.893 0.947DF0.8870.0570.8890.8870.887 0.956 EF 0.92 0.04 0.921 0.92 0.92 0.979 AM0.7930.1030.7930.7930.7930.923BM0.8930.0530.8970.8930.8940.962CM 0.967 0.017 0.967 0.967 0.967 0.995 DM 0.967 0.017 0.967 0.967 0.967 0.992EM 0.967 0.017 0.967 0.987 0.987 1.0 AF0.7330.1330.7350.7330.7340.883BF0.7670.1170.7660.7670.7660.898CF0.8870.0570.8880.8870.8870.951DF 0.900 0.05 0.901 0.9 0.9 0.964 EF 0.913 0.043 0.914 0.913 0.913 0.983 AM0.7730.1130.7750.7730.7730.871BM0.9130.0430.9150.9130.9140.947CM0.9670.0170.9670.9670.9670.978DM 0.973 0.013 0.974 0.973 0.973 0.976 EM 0.993 0.003 0.993 0.993 0.993 0.994 AF0.7530.1230.7530.7530.7530.866BF0.8070.0970.8050.8070.8050.904CF 0.900 0.050 0.900 0.900 0.900 0.937 DF0.8930.0530.8930.8930.8930.943EF 0.907 0.047 0.909 0.907 0.907 0.94 AM0.7730.1130.7750.7730.7720.83BM0.820.090.8230.820.8230.865CM0.860.070.8580.860.8570.895DM 0.933 0.033 0.934 0.933 0.933 0.95 EM 0.953 0.023 0.953 0.953 0.953 0.965 AF0.7330.1330.7520.7330.7340.8BF0.7200.140.7580.720.7130.79CF0.6670.1670.7150.6670.6080.75DF 0.860 0.07 0.862 0.86 0.859 0.895 EF 0.92 0.04 0.922 0.92 0.92 0.94 AM0.8930.0530.8950.8930.8930.944BM0.9670.0170.9670.9670.9670.982CM0.9670.0170.9670.9670.9670.983DM 0.973 0.013 0.974 0.973 0.973 0.985 EM 0.973 0.013 0.973 0.973 0.973 0.985 AF0.8670.0670.8680.8670.8670.922BF0.9070.0470.9070.9070.9070.947CF0.9070.0470.9070.9070.9070.943DF 0.933 0.033 0.934 0.933 0.934 0.965 EF 0.953 0.023 0.954 0.953 0.953 0.97 M(20长度特征)80.00±1.8389.33±1.0479.30±1.62M(195长度特征)83.33±2.2190.00±1.0689.33±0.69M(250角度特征)92.00±1.0596.00±0.5596.70±0.85M(400角度特征)90.00±1.1196.70±0.4796.70±0.28M(51混合特征) 96.00 ± 0.55 97.33 ± 0.21 98.67 ± 0.53 F(20长度特征)72.67±2.3177.33±1.4473.33±1.94F(195长度特征)77.33±1.5175.33±1.2176.67±1.20F(250角度特征)81.33±2.7889.33±0.9588.67±0.47F(400角度特征)76.67±2.5188.67±0.5590.00±0.38F(51混合特征) 81.33 ± 2.10 92.00 ± 0.35 91.33 ± 0.32 M(20长度特征)77.33±2.17 89.33 ± 1.65 77.33±1.03M(195长度特征)91.33±0.95 96.67 ± 0.54 82.00±0.99M(250角度特征)96.70±0.85 96.70 ± 0.56 86.00±0.32M(400角度特征)97.30±0.35 97.30 ± 0.61 93.30±0.52M(51混合特征) 99.33 ± 1.25 97.33 ± 0.49 95.33 ± 0.49 F(20长度特征)75.33±2.8786.67±1.1973.33±1.67F(195长度特征)80.67±1.1490.67±1.2972.00±1.43F(250角度特征)90.00±1.1090.67±0.8866.67±1.08F(400角度特征)89.33±0.8593.33±1.3086.00±0.73F(51混合特征) 90.67 ± 0.94 95.33 ± 0.76 92.00 ± 0.89 段晓东, 王存睿, 刘向东, 刘慧.人脸的民族特征抽取及其识别.计算机科学, 2010, 37(8):276-279, 301 http://www.docin.com/p-1301615906.html

Duan Xiao-Dong, Wang Cun-Rui, Liu Xiang-Dong, Liu Hui. Minorities features extraction and recognition of human faces. Computer Science , 2010, 37(8):276-279, 301 http://www.docin.com/p-1301615906.html Li Z D, Duan X D, Zhang Q L. A novel survey based on multiethnic facial semantic web. TELKOMNIKA , 2013, 11(9):5076-5083 https://www.researchgate.net/profile/Qingling_Zhang4/publication/275415292_A_Novel_Survey_Based_on_Multiethnic_Facial_Semantic_Web/links/5675798408aebcdda0e46b5e.pdf?inViewer=0&pdfJsDownload=0&origin=publication_detail 段晓东, 李泽东, 王存睿, 张庆灵, 刘晓东.基于AFS的多民族人脸语义描述与挖掘方法研究.计算机学报, 2016, 39(7):1435-1449 doi: 10.11897/SP.J.1016.2016.01435

Duan Xiao-Dong, Li Ze-Dong, Wang Cun-Rui, Zhang Qing-Ling, Liu Xiao-Dong. Multi-ethnic face semantic description and mining method based on AFS. Chinese Journal of Computer , 2016, 39(7):1435-1449 doi: 10.11897/SP.J.1016.2016.01435 詹德川, 周志华.基于集成的流形学习可视化.计算机研究与发展, 2005, 42(9):1533-1537 http://www.cnki.com.cn/Article/CJFDTOTAL-WXJY201508008.htm

Zhan De-Chuan, Zhou Zhi-Hua. Ensemble-based manifold learning for visualization. Journal of Computer Research and Development , 2005, 42(9):1533-1537 http://www.cnki.com.cn/Article/CJFDTOTAL-WXJY201508008.htm 何力, 张军平, 周志华.基于放大因子和延伸方向研究流形学习算法.计算机学报, 2005, 28(12):2000-2009 doi: 10.3321/j.issn:0254-4164.2005.12.007

He Li, Zhang Jun-Ping, Zhou Zhi-Hua. Investigating manifold learning algorithms based on magnification factors and principal spread directions. Chinese Journal of Computers , 2005, 28(12):2000-2009 doi: 10.3321/j.issn:0254-4164.2005.12.007 曾宪华, 罗四维.动态增殖流形学习算法.计算机研究与发展, 2007, 44(9):1462-1468 https://www.wenkuxiazai.com/doc/d9249222a32d7375a417801c.html

Zeng Xian-Hua, Luo Si-Wei. A dynamically incremental manifold learning algorithm. Journal of Computer Research and Development , 2007, 44(9):1462-1468 https://www.wenkuxiazai.com/doc/d9249222a32d7375a417801c.html 张大明, 符茂胜, 罗斌.基于二维近邻保持嵌入的图像识别.模式识别与人工智能, 2011, 24(6):810-815 http://www.oalib.com/paper/4238112

Zhang Da-Ming, Fu Mao-Sheng, Luo Bin. Image recognition with two-dimensional neighbourhood preserving embedding. Pattern Recognition and Artificial Intelligence , 2011, 24(6):810-815 http://www.oalib.com/paper/4238112 Li Z D, Zhang Q L, Duan X D, Wei W. Semantic knowledge based on fuzzy system for describing facial expression. In: Proceedings of the 2017 Chinese Control Conference. 2017: 9865-9870 http://www.researchgate.net/publication/320114803_Semantic_knowledge_based_on_fuzzy_system_for_describing_facial_expression 李小林. 中国南方五个少数民族人群面部五官形态特征研究[硕士学位论文], 重庆医科大学, 中国, 2008 http://www.wanfangdata.com.cn/details/detail.do?_type=degree&id=Y1353533

Li Xiao-Lin. Analysis of Morphous Characteristics of Facial Reconstruction and the Five Organs in Chinese South Five National Minorities Crowd[Master dissertation], Chongqing Medical University, China, 2008 http://www.wanfangdata.com.cn/details/detail.do?_type=degree&id=Y1353533 赵明, 文建军, 赵霞, 孙嬿嬿.藏、蒙、土、回、汉五民族青年面部测量数据分析.青海医学院学报, 1995, 16(3):7-8 http://paper.usc.cuhk.edu.hk/Details.aspx?id=5247

Zhao Ming, Wen Jian-Jun, Zhao Xia, Sun Yan-Yan. An analysis on measuring data of face in 169 young students of Tibetan, Mongolian, Tu, Huis, and Han nationality. Journal of Qinghai Medical College , 1995, 16(3):7-8 http://paper.usc.cuhk.edu.hk/Details.aspx?id=5247 王朝云, 蒋刚毅, 郁梅, 陈芬.基于流形特征相似度的感知图像质量评价.自动化学报, 2016, 42(7):1113-1124 http://www.aas.net.cn/CN/Y2016/V42/I7/1113

Wang Chao-Yun, Jiang Gang-Yi, Yu Mei, Chen Fen. Manifold feature similarity based perceptual image quality assessment. Acta Automatica Sinica , 2016, 42(7):1113-1124 http://www.aas.net.cn/CN/Y2016/V42/I7/1113 洪泉, 陈松灿, 倪雪蕾.子模式典型相关分析及其在人脸识别中的应用.自动化学报, 2008, 34(1):21-30 http://www.aas.net.cn/CN/abstract/abstract16043.shtml

Hong Quan, Chen Song-Can, Ni Xue-Lei. Sub-pattern canonical correlation analysis with application in face recognition. Acta Automatica Sinica , 2008, 34(1):21-30 http://www.aas.net.cn/CN/abstract/abstract16043.shtml 高全学, 潘泉, 梁彦, 张洪才, 程咏梅.基于描述特征的人脸识别研究.自动化学报, 2006, 32(3):386-392 http://www.aas.net.cn/CN/abstract/abstract15824.shtml

Gao Quan-Xue, Pan Quan, Liang Yan, Zhang Hong-Cai, Cheng Yong-Mei. Face recognition based on expressive features. Acta Automatica Sinica , 2006, 32(3):386-392 http://www.aas.net.cn/CN/abstract/abstract15824.shtml

地址：北京中关村东路95号邮政编码：100190 E-mail： [email protected]

电话：010-82544677 (日常咨询和稿件处理)， 010-82544653(费用管理、寄刊)

北京仁和汇智信息技术有限公司开发技术支持： [email protected]