特征选择对FHMM性能影响研究

doi:10.3969/j.issn.1001-2400.2010.05.028

J4 ›› 2010, Vol. 37 ›› Issue (5): 934-940.doi: 10.3969/j.issn.1001-2400.2010.05.028

特征选择对FHMM性能影响研究

陈昌红^1，2;赵恒¹;梁继民¹;焦李成³

(1. 西安电子科技大学生命科学与技术学院生命科学研究中心，陕西西安 710071；
2. 南京邮电大学通信与信息工程学院，江苏南京 210003；

收稿日期:2010-03-10 出版日期:2010-10-20 发布日期:2010-10-11
通讯作者: 陈昌红
作者简介:陈昌红(1982-)，女，博士，E-mail: chhchen127@gmail.com．
基金资助:
国家重点基础研究发展计划973资助项目(2006CB705700);教育部创新团队计划资助项目(IRT0645);国家自然科学基金资助项目(60902038,60872154);陕西省自然科学基础研究计划资助项目(SJ08F18)

Influence of feature selection on FHMM

CHEN Chang-hong^1,2;ZHAO Heng¹;LIANG Ji-min¹;JIAO Li-cheng³

(1. Research Center of Life Science, School of Life Sciences and Technolog, Xidian Univ., Xi'an 710071， China;
2. College of Communication and Information Eng., Nanjing Univ. of Posts and Telecommunications, Nanjing 210003， China;

Received:2010-03-10 Online:2010-10-20 Published:2010-10-11
Contact: CHEN Chang-hong

摘要/Abstract

摘要：

在利用因子隐马尔可夫模型(Factorial Hidden Markov Model， FHMM)进行分类识别的过程中，特征选择是影响其性能的主要因素．通过研究特征选择对FHMM性能的影响，提出了一种性能分析的方案，得出了选择FHMM特征的准则．将FHMM引入到步态识别中，提取4种步态特征，得到使用不同特征组合的FHMM的实验结果．使用McNemar检验的方法将其与单个特征的识别性能做比较，结合由正则典型相关分析得到的维数不同的特征间的相关性，分析得到以下结论: 基于FHMM的识别性能与特征间的相关性并没有必然联系，其性能更多地受到特征间的识别性能差异和单个特征的识别性能的影响．为发挥FHMM的优越性，应选择特征间识别性能差异小和单个特征识别性能好的特征组合，在此基础上特征间相关性越小越好．

关键词: 因子隐马尔可夫模型, 特征选择, 正则典型相关分析, McNemar检验, 步态识别

Abstract:

In the application of the factorial hidden Markov model (FHMM) to classification and identification, feature selection is the major factor impacting the performance of FHMM. An analytical scheme is proposed based on the impact of feature selection on the performance of FHMM and a feature selection criterion for FHMM is obtained. FHMM is introduced into gait recognition. Four gait features are extracted and the experimental results of FHMM with different feature combinations are gained. McNemar's test is employed to estimate the performance of FHMM and HMM with a single feature and regularized Canonical Correlation Analysis is used to calculate the relativity between features with different dimensions. Combining the experimental results and the relativities leads to the following conclusion: the recognition performance of FHMM does not have positive connection with the features' relativity, which is greatly influenced by the performance diversity between features and the performance of recognition with the single feature. In order to make good use of FHMM's advantage, the features should be selected according to the small performance diversity between features and the good performance of the single feature. Based on this, the lower the features' relativity, the better.

Key words: factorial hidden Markov model, feature selection, regularized canonical correlation analysis, McNemar's test, gait recognition

陈昌红;赵恒;梁继民;焦李成. 特征选择对FHMM性能影响研究[J]. J4, 2010, 37(5): 934-940.

CHEN Chang-hong;ZHAO Heng;LIANG Ji-min;JIAO Li-cheng. Influence of feature selection on FHMM[J]. J4, 2010, 37(5): 934-940.

参考文献

[1] Ghahramani Z, Jordan M I. Factorial Hidden Markov Models [J] . Machine learning, 1997, 29(2): 245-273.
[2] Baum L E, Eagon J A. An Inequality with Applications to Statistical Estimation for Probabilistic Functions of Markov Processes and to a Model for Ecology [J] . Bull Amer Math Soc, 1967(73): 360-363.
[3] Logan B, Moreno P J. Factorial Hidden Markov Models for Speech Recognition: Preliminary Experiments [C] //Technical Report Series CFL 97／7. Cambridge: Cambridge Research Laboratory, 1997.
[4] Jacobs R A, Jiang W, Tanner M A. Factorial Hidden Markov Models and the Generalized Backfitting Algorithm [J] . Neural computation, 2002, 14(10): 2415-2437.
[5] Reyes-Gomez M J, Raj B, Ellis D. Multi-channel Source Separation by Factorial HMMs [C] //Proc ICASSP. Hong Kong: IEEE, 2003: 1664-1667.
[6] Deoras A N, Hasegawa-Johnson M. A Factorial HMM Approach to Simultaneous Recognition of Isolated Digits Spoken by Multiple Talkers on One Audio Channel [C] //Proc ICASSP. Montreal: IEEE, 2004: 1861-1864.
[7] Kulic D, Takano W, Nakamura Y. Representability of Human Motions by Factorial Hidden Markov Models [C] //Proc IROS. Tokyo: IEEE, 2007: 2388-2393.
[8] Lee D, Kulic D, Nakamura Y. Missing Motion Data Recovery Using Factorial Hidden Markov Models [C] //Proc ICRA. Tokyo: IEEE, 2008: 1722-1728.
[9] Chen C, Liang J, Zhao H, et al. Factorial HMM and Parallel HMM for Gait Recognition [J] . IEEE Trans on Systems, Man and Cybernetics-Part C: Applications and Reviews, 2009, 39(1): 114-123.
[10] Duh K. Jointly Labeling Multiple Sequences: a Factorial Hmm Approach [C] //Proc of the Association for Computational Lingustics Morristown. Ann Arbor: Student Research Workshop, 2005: 19-24.
[11] Gael J V, Teh Y W, Ghahramani Z. The Infinite Factorial Hidden Markov Model [J] . Advances in Neural Information Processing Systems, 2008(21):1697-1704.
[12] Rodgers J L, Nicewander W A. Thirteen Ways to Look at the Correlation Coefficient [J] . The American Statistician, 1988(42): 59-66.
[13] Zhang S, Xie L, Adams M D. Entropy Based Feature Selection Scheme for Real Time Simultaneous Localization and Map Building [C] //Proc IRS/RSJ. Edmonton: IEEE, 2005: 1175-1180.
[14] Akaike H. Canonical Correlation Analysis of Time Series and the Use of an Information Criterion [J] . System Identification: Advances and Case Studies, 1976, (126): 27-96.
[15] Gonzalez I, Dejean S, Martin P G P, et al. CCA: an R Package to Extend Canonical Correlation Analysis [J] . Journal of Statistical Software, 2007, 23(12): 1-14.
[16] Chen C, Liang J, Zhao H, et al. Gait Recognition Using Hidden Markov Model [C] //Advances in Natural Computation. Xi'an: Springer Verlag, 2006: 399-407.
[17] Leurgans S E, Moyeed R A, Silverman B W. Canonical Correlation Analysis When the Data Are Curves[J] . Journal of the Royal Statistical Society B, 1993, 55(3): 725-740.
[18] Liu Y, Collins R T, Tsin Y. Gait Sequence Analysis Using Frieze Patterns [C] //Proc ECCV. London: Springer, 2002: 657-671.
[19] Teague M R. Image Analysis Via the General Theory of Moments [J] . Journal of the Optical Society of America, 1979, 70(8): 920-930.
[20] Gross R, Shi J. The Cmu Motion of Body (Mobo) Database [C] //Technical Report CMU-RI-TR-01-18. Pittsburgh: Robotics Institute, Carnegie Mellon University, 2001.
[21] CASIA Gait Database [DB/OL] . [2010-02-05] . http://www.cbsr.ia.ac.cn／english／Gait％20Databases.asp.
[22] Clark A F, Clark C. Performance Characterization in Computer Vision a Tutorial [DB/OL] . [2010-02-10] . http://peipa.esser.ac.uk／benchmark／tutorial／esser／tutorial.pdf.

特征选择对FHMM性能影响研究

Influence of feature selection on FHMM

PDF (PC)

赞

可视化

被引次数

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 14

Metrics

本文评价

推荐阅读 10

[1]	戚艳军,孔月萍,王佳婧,朱旭东. 一种LSTM与CNN相结合的步态识别方法[J]. 西安电子科技大学学报, 2021, 48(5): 78-85.
[2]	许缓缓,李洪梅,李富余,孙学梅. 一种时空卷积的步态识别方法[J]. 西安电子科技大学学报, 2021, 48(4): 144-150.
[3]	刘景美,高源伯. 自适应分箱特征选择的快速网络入侵检测系统[J]. 西安电子科技大学学报, 2021, 48(1): 176-182.
[4]	肖利军,郭继昌,顾翔元. 一种采用冗余性动态权重的特征选择算法[J]. 西安电子科技大学学报, 2019, 46(5): 155-161.
[5]	李海霞;张擎;王青;尹义龙;郝凡昌. 肺部CT图像特征的设备无关性研究[J]. J4, 2016, 43(1): 157-161+179.
[6]	李志鹏;马田香;杜兰;徐丹蕾;刘宏伟;张子敬. 在雷达HRRP识别中多特征融合多类分类器设计[J]. J4, 2013, 40(1): 111-117.
[7]	豆增发;高琳. 利用膜粒子群优化的条件随机域特征选择[J]. J4, 2012, 39(5): 107-112.
[8]	李雪;田斌;詹毅;孙永军. 一种基于ITD参数的数字调制识别算法[J]. J4, 2010, 37(6): 1132-1136.
[9]	尚荣华;焦李成;吴建设;马文萍;李阳阳. 用于非监督特征选择的免疫克隆多目标优化算法[J]. J4, 2010, 37(1): 18-22.
[10]	陆满君;詹毅;司锡才;杨小牛. 通信辐射源瞬态特征提取和个体识别方法[J]. J4, 2009, 36(4): 736-740.
[11]	朱虎明;焦李成. 并行免疫克隆特征选择算法 [J]. J4, 2008, 35(5): 853-857.
[12]	王勇1;2;陶晓玲1. 分级结构的AdaBoost入侵检测方法研究 [J]. J4, 2008, 35(2): 345-350.
[13]	陈实1;田玉敏1;黄万红2;高有行1. 用时变模板自动识别行人的步态 [J]. J4, 2007, 34(4): 605-610.
[14]	张军英1;2;刘利平1. 基于部分独立分量分析的盲源分离[J]. J4, 2004, 31(3): 334-337.