正在加载图片...
·510· 智能系统学报 第8卷 DA的数据参数设定在一定范围内对性别识别的结 sis and Machine Intelligence,2002,24(5):707-711. 果影响较小.在16个块大小和121个块大小的条件 [7]SHAKHNAROVICH G,VIOLA P,MOGHADDAM B.A u- 下,本文算法的识别率分别维持在94%左右和95% nified learning framework for real time face detection and 左右.只在全局特征数据的基础上,增加10~20维 classification[C]//IEEE Conf on Automatic Face and Ges- 数据,即可实现较高的识别率,验证了本文提出的特 ture Recognition.Washinton,DC,USA,2002:14-21. [8]COSTEN N P,BROWN M,AKAMATSU S.Sparse models 征融合的有效性, for gender classification[C]//IEEE Conf on Automatic Face 本文算法涉及的可调参数分别为图像分块大 and Gesture Recognition.Seoul,Korea,2004:201-206. 小、PCA特征提取数据维数和OLDA特征提取数据 [9]SAATCI Y,TOWN C.Cascaded classification of gender and 维数,图4表明这些可调参数在相当大的调试范围 facial expression using active appearance models [C]// 内对本文算法的识别率均影响较小,表现了该方法 IEEE Conf on Automatic Face and Gesture Recognition. 具有较强的稳定性。 Southampton,UK,2006:393-398. [10]BRUNELLI R,POGGIO T.HyperBF networks for gender 6结束语 classification[C]//Defense Advanced Research Projects 本文以单样本人脸图像研究为起点,建立了相 Agency Image Understanding Workshop.San Diego,USA, 1992:311-314. 互独立的训练库和测试库,改进了LDA算法,使之 [11]EDELMAN B,VALENTIN D,ABDI H.Sex classification 能够应用于二分类问题,打破了传统LDA存在的秩 of face areas:how well can a linear neural network predict 问题.提出PCA+OLDA的全局特征提取方法,使得 human performance[J.Joural of Biological System, 图像特征中性别信息比重增加:采用图像全局特征 1998,6(3):241-264. 和局部特征融合的方式,获得图像的性别特征,使得 [12]文益民,吕宝粮.最小最大模块化支持向量机改进研究 性别特征的描述更加全面.实验证明,本文方法对各 [J].计算机工程与应用,2005,41(19):185-188. 种可调参数具有鲁棒性,从一定程度上削弱了参数 WEN Yimin,LO Baoliang.Improvement research of min-m 多变给性别识别带来的影响.人脸图像易于采集,且 ax modular support vector machine[J].Journal of Comput- 易被人接受,训练样本数量增加,使得性别识别模型 er Engineering and Applications,2005,41(19):185- 188. 的训练更加充分可靠,实用性明显增强. [13]孙宁,冀贞海,邹采荣,等.基于局部二元模式算子的人 参考文献: 脸性别分类方法[J].华中科技大学学报:自然科学版, 2007,35(z1):177-181. [1]GAO W,CHAO B,SHAN S,et al.The CAS-PEAL large- SUN Ning,JI Zhenhai,ZOU Cairong,et al.Gender clas- scale Chinese face database and baseline evaluations[J]. sification based on local binary pattern[J].Journal of Hua- IEEE Transactions on System,Man,and Cybernetics,Part zhong University of Science and Technology,2007,35 A:Systems and Human,2008,38(1):149-161. (z1):177-181 [2]XIA Bin,SUN He,LO Baoliang.Multiview gender classifi- [14]0ZBUDAK O,KIRC M,CAKIR Y,et al.Effects of the cation based on local Gabor binary mapping pattern and sup- facial and racial features on gender classification [C]// port vector machines[C]//International Joint Conference on IEEE Mediterranean Electrotechnical Conference.Valetta, Neural Networks.Hong Kong,China,2008:3388-3395. Malta,2010:26-29. [3]GOLOMB B A,LAWRENCE D T,SEJNOWSKI T J.SEX- [15]YANG T,KECMAN V.Face recognition with adaptive lo- NET:a neural network identifies sex from human face cal hyperplane algorithm[J].Pattern Anal Applic,2010, [C]//Advances in Neural Information Processing Systems. 13(1):79-83. 「S.L.1,USA,1991:572-577. [16]YE J.Characterization of a family of algorithms for general- [4]COTTRELL G W,METCALFE J.EMPATH:face,emotion ized discriminant analysis on undersampled problemsJ and gender recognition using holons[C]//Advances in Neu- Journal of Machine Learning Research,2005,6(4):483- ral Information Processing Systems.[S.1.],USA,1991: 502. 564-771. [17]OJALA T,PIETIKAINEN M,MAENPAA T.Multiresolu- [5]TAMURA S H,MITSUMOTO K H.Male/female identifica- tion gray-scale and rotation invariant texture classification tion from 8x6 very low resolution face images by neural net- with local binary pattern[J].IEEE Transactions on Pattern work[J].Pattern Recognition,1996,29(2):331-335. Analysis and Machine Intelligence,2002,24(7):971- 6]MOGHADDAM B.YANG M H.Gender classification with 986. support vector Machines[J].IEEE Trans on Pattern Analy- [18]WANG Yu,MU Zhichun,ZENG Hui.Block-based andDA 的数据参数设定在一定范围内对性别识别的结 果影响较小.在 16 个块大小和 121 个块大小的条件 下,本文算法的识别率分别维持在 94%左右和 95% 左右.只在全局特征数据的基础上,增加 10 ~ 20 维 数据,即可实现较高的识别率,验证了本文提出的特 征融合的有效性. 本文算法涉及的可调参数分别为图像分块大 小、PCA 特征提取数据维数和 OLDA 特征提取数据 维数,图 4 表明这些可调参数在相当大的调试范围 内对本文算法的识别率均影响较小,表现了该方法 具有较强的稳定性. 6 结束语 本文以单样本人脸图像研究为起点,建立了相 互独立的训练库和测试库,改进了 LDA 算法,使之 能够应用于二分类问题,打破了传统 LDA 存在的秩 问题.提出 PCA+OLDA 的全局特征提取方法,使得 图像特征中性别信息比重增加;采用图像全局特征 和局部特征融合的方式,获得图像的性别特征,使得 性别特征的描述更加全面.实验证明,本文方法对各 种可调参数具有鲁棒性,从一定程度上削弱了参数 多变给性别识别带来的影响.人脸图像易于采集,且 易被人接受,训练样本数量增加,使得性别识别模型 的训练更加充分可靠,实用性明显增强. 参考文献: [1]GAO W, CHAO B, SHAN S, et al. The CAS⁃PEAL large⁃ scale Chinese face database and baseline evaluations [ J]. IEEE Transactions on System, Man, and Cybernetics, Part A: Systems and Human, 2008, 38(1): 149⁃161. [2]XIA Bin, SUN He, LÜ Baoliang. Multiview gender classifi⁃ cation based on local Gabor binary mapping pattern and sup⁃ port vector machines[C] / / International Joint Conference on Neural Networks. Hong Kong, China, 2008: 3388⁃3395. [3]GOLOMB B A, LAWRENCE D T, SEJNOWSKI T J. SEX⁃ NET: a neural network identifies sex from human face [C] / / Advances in Neural Information Processing Systems. [S.l.], USA, 1991: 572⁃577. [4]COTTRELL G W, METCALFE J. EMPATH: face, emotion and gender recognition using holons[C] / / Advances in Neu⁃ ral Information Processing Systems. [ S. l.], USA, 1991: 564⁃771. [5]TAMURA S H, MITSUMOTO K H. Male / female identifica⁃ tion from 8×6 very low resolution face images by neural net⁃ work[J]. Pattern Recognition, 1996, 29(2): 331⁃335. [6] MOGHADDAM B, YANG M H. Gender classification with support vector Machines[J]. IEEE Trans on Pattern Analy⁃ sis and Machine Intelligence, 2002, 24(5): 707⁃711. [7]SHAKHNAROVICH G, VIOLA P, MOGHADDAM B. A u⁃ nified learning framework for real time face detection and classification[C] / / IEEE Conf on Automatic Face and Ges⁃ ture Recognition. Washinton, DC, USA, 2002: 14⁃21. [8]COSTEN N P, BROWN M, AKAMATSU S. Sparse models for gender classification[C] / / IEEE Conf on Automatic Face and Gesture Recognition. Seoul, Korea, 2004: 201⁃206. [9]SAATCI Y, TOWN C. Cascaded classification of gender and facial expression using active appearance models [ C] / / IEEE Conf on Automatic Face and Gesture Recognition. Southampton, UK, 2006: 393⁃398. [10]BRUNELLI R, POGGIO T. HyperBF networks for gender classification[ C] / / Defense Advanced Research Projects Agency Image Understanding Workshop. San Diego, USA, 1992: 311⁃314. [11]EDELMAN B, VALENTIN D, ABDI H. Sex classification of face areas: how well can a linear neural network predict human performance [ J ]. Journal of Biological System, 1998, 6(3): 241⁃264. [12]文益民,吕宝粮. 最小最大模块化支持向量机改进研究 [J]. 计算机工程与应用, 2005, 41(19): 185⁃188. WEN Yimin, LÜ Baoliang. Improvement research of min⁃m ax modular support vector machine[J]. Journal of Comput⁃ er Engineering and Applications, 2005, 41 ( 19): 185⁃ 188. [13]孙宁,冀贞海,邹采荣,等. 基于局部二元模式算子的人 脸性别分类方法[J]. 华中科技大学学报:自然科学版, 2007, 35(z1): 177-181. SUN Ning, JI Zhenhai, ZOU Cairong, et al. Gender clas⁃ sification based on local binary pattern[J]. Journal of Hua⁃ zhong University of Science and Technology, 2007, 35 (z1): 177-181. [14]ÖZBUDAK O, KIRCı M, CAKIR Y, et al. Effects of the facial and racial features on gender classification [ C] / / IEEE Mediterranean Electrotechnical Conference. Valetta, Malta, 2010: 26⁃29. [15]YANG T, KECMAN V. Face recognition with adaptive lo⁃ cal hyperplane algorithm[ J]. Pattern Anal Applic, 2010, 13(1): 79⁃83. [16]YE J. Characterization of a family of algorithms for general⁃ ized discriminant analysis on undersampled problems[ J]. Journal of Machine Learning Research, 2005, 6(4): 483⁃ 502. [17]OJALA T, PIETIKAINEN M, MAENPAA T. Multiresolu⁃ tion gray⁃scale and rotation invariant texture classification with local binary pattern[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24 ( 7): 971⁃ 986. [18] WANG Yu, MU Zhichun, ZENG Hui. Block⁃based and ·510· 智 能 系 统 学 报 第 8 卷
<<向上翻页向下翻页>>
©2008-现在 cucdc.com 高等教育资讯网 版权所有