Fast approximation of Kullback-Leibler distance for dependence trees and hidden Markov models

被引:135
作者
Do, MN [1 ]
机构
[1] Univ Illinois, Dept Elect & Comp Engn, Coordinated Sci Lab, Urbana, IL 61801 USA
[2] Univ Illinois, Beckman Inst, Urbana, IL 61801 USA
关键词
dependence tree models; hidden Markov models; Kullback-Leibler distance;
D O I
10.1109/LSP.2003.809034
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
We present a fast algorithm to approximate the Kullback-Leibler distance (KLD) between two dependence tree models. The algorithm uses the "upward" (or "forward") procedure to compute an upper bound for the KLD. For hidden Markov models, this algorithm is reduced to a simple expression. Numerical experiments show that for a similar accuracy, the proposed algorithm offers a saving of hundreds of times in computational complexity compared to the commonly used Monte Carlo method. This makes the proposed algorithm important for real-time applications, such as image retrieval.
引用
收藏
页码:115 / 118
页数:4
相关论文
共 7 条
[1]  
Cover T. M., 2005, ELEM INF THEORY, DOI 10.1002/047174882X
[2]   Wavelet-based statistical signal processing using hidden Markov models [J].
Crouse, MS ;
Nowak, RD ;
Baraniuk, RG .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 1998, 46 (04) :886-902
[3]   Rotation invariant texture characterization and retrieval using steerable wavelet-domain hidden Markov models [J].
Do, MN ;
Vetterli, M .
IEEE TRANSACTIONS ON MULTIMEDIA, 2002, 4 (04) :517-527
[4]   A PROBABILISTIC DISTANCE MEASURE FOR HIDDEN MARKOV-MODELS [J].
JUANG, BH ;
RABINER, LR .
AT&T TECHNICAL JOURNAL, 1985, 64 (02) :391-408
[5]   A TUTORIAL ON HIDDEN MARKOV-MODELS AND SELECTED APPLICATIONS IN SPEECH RECOGNITION [J].
RABINER, LR .
PROCEEDINGS OF THE IEEE, 1989, 77 (02) :257-286
[6]   PARAMETER-ESTIMATION OF DEPENDENCE TREE MODELS USING THE EM ALGORITHM [J].
RONEN, O ;
ROHLICEK, JR ;
OSTENDORF, M .
IEEE SIGNAL PROCESSING LETTERS, 1995, 2 (08) :157-159
[7]  
SINGER Y, 1998, P ADV NEURAL INFORMA, V11, P578