|Table of Contents|

[1] Yu Lu, Wu Lenan, Xie Jun, et al. Optimal state and branch sequence based parameter estimationof continuous hidden Markov model [J]. Journal of Southeast University (English Edition), 2005, 21 (2): 136-140. [doi:10.3969/j.issn.1003-7985.2005.02.004]
Copy

Optimal state and branch sequence based parameter estimationof continuous hidden Markov model()
基于最优状态和分支序列的连续隐Markov模型参数估计
Share：

Journal of Southeast University (English Edition)[ISSN:1003-7985/CN:32-1325/N]

Volumn:
21
Issue:
2005 2
Page:
136-140
Research Field:
Computer Science and Engineering
Publishing date:
2005-06-30

Info

Title:
Optimal state and branch sequence based parameter estimationof continuous hidden Markov model
基于最优状态和分支序列的连续隐Markov模型参数估计
Author(s):
Yu Lu1, 2, Wu Lenan1, Xie Jun3
1Department of Radio Engineering, Southeast University, Nanjing 210096, China
2 Institute of Communications Engineering, PLA University of Science and Technology, Nanjing 210007, China
3 Institute of Command Automation, PLA University of Science and Technology, Nanjing 210007, China
俞璐1, 2, 吴乐南1, 谢钧3
1 东南大学无线电工程系, 南京 210096; 2 解放军理工大学通信工程学院, 南京 210007; 3 解放军理工大学指挥自动化学院, 南京 210007
Keywords:
continuous hidden Markov model;  optimal state and branch sequence;  maximum likelihood;  convergence;  Viterbi algorithm
连续隐Markov模型;  最优状态和分支序列;  最大似然;  收敛性;  Viterbi算法
PACS:
TP391
DOI:
10.3969/j.issn.1003-7985.2005.02.004
Abstract:
A parameter estimation algorithm of the continuous hidden Markov model is introduced and the rigorous proof of its convergence is also included.The algorithm uses the Viterbi algorithm instead of K-means clustering used in the segmental K-means algorithm to determine optimal state and branch sequences.Based on the optimal sequence, parameters are estimated with maximum-likelihood as objective functions.Comparisons with the traditional Baum-Welch and segmental K-means algorithms on various aspects, such as optimal objectives and fundamentals, are made.All three algorithms are applied to face recognition.Results indicate that the proposed algorithm can reduce training time with comparable recognition rate and it is least sensitive to the training set.So its average performance exceeds the other two.
提出了一种连续隐Markov模型参数估计算法, 并利用全局收敛定理严格证明了算法的收敛性.该算法用Viterbi算法取代分段K平均算法中的聚类方法, 直接确定出最优状态和分支序列, 并依据最优序列以最大似然为优化准则进行参数估计.阐述了该算法与Baum-Welch和分段K平均2种经典算法在目标函数、优化准则和工作原理等方面的关系, 并将3种算法应用于人脸识别.实验结果表明, 该算法在获得相当识别率的同时缩短了训练时间, 并降低了识别结果对训练样本集的敏感性, 在3种算法中总体性能最优.

References:

[1] Rabiner L R.A tutorial on hidden Markov models and selected applications in speech recognition [J].Proceedings of the IEEE, 1989, 77(2):257-285.
[2] Juang B H, Rabiner L R.The segmental K-means algorithm for estimating parameters of hidden Markov models [J].IEEE Trans on Acoustics Speech and Signal Processing, 1990, 38(9):1639-1641.
[3] Andrieu C, Doucet A.Simulated annealing for maximum a posteriori parameter estimation of hidden Markov models [J].IEEE Trans on Information Theory, 2000, 46(5):994-1004.
[4] Yishai A B, Burshtein D.A discriminative training algorithm for hidden Markov models [J].IEEE Trans on Speech and Audio Processing, 2004, 12(5):204-217.
[5] Juang B H, Levinson S E, Sondhi M M.Maximum likelihood estimation for multivariate mixture observations of Markov chains [J].IEEE Trans on Information Theory, 1986, 32(3):307-309.
[6] Rabiner L R, Juang B H, Levinson S E.Some properties of continuous hidden Markov model representations [J].AT& T Technical Journal, 1985, 64(6):1251-1269.
[7] Wu C F Jeff.On the convergence properties of the EM algorithm [J].The Annals of Statistics, 1983, 11(1):95-103.
[8] Chen Baolin.Optimization theory and algorithm [M].Beijing:Tsinghua University Press, 1989.287-299.(in Chinese)
[9] Kohir V V, Desai U B.Face recognition [A].In:Proceedings of the IEEE Symposium on Circuits and Systems [C].Geneva, Switzerland:Presses Polytechniques et Universitaires Romandes, 2000, 5:305-308.
[10] Bicego M, Castellani U, Murino V.Using hidden Markov models and wavelets for face recognition [A].In:Wemer Bob, ed.Proceedings of the 12th International Conference on Image Analysis and Processing [C].Mantova, Italy:IEEE Computer Society, 2003.52-56.

Memo

Memo:
Biographies: Yu Lu(1973—), female, graduate;Wu Lenan(corresponding author), male, doctor, professor, wuln@seu.edu.cn.
Last Update: 2005-06-20