Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle

Kenneth Rose, David Jonathan Miller, Allen Gersho

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

We address the variable rate tree-structured vector quantizer design problem, wherein the rate is measured by the quantizer's entropy. For this problem, tree pruning via the Generalized Breiman-Friedman-Olshen-Stone algorithm obtains solutions which are optimal over the restricted solution space consisting of all pruned trees derivable from an initial tree. We develop a joint optimization method which is inspired by the deterministic annealing algorithm for data clustering, and which extends our previous work on tree-structured vector quantization. The method is based on the principle of minimum cross entropy, using informative priors to approximate the unstructured solution while imposing the structural constraint. As in the original deterministic annealing method, the number of distinct codevectors (and hence the tree) grows by a sequence of bifurcations in the process, which occur as solutions of a free energy minimization. Our method obtains performance gains over growing and pruning methods for variable rate quantization of Gauss-Markov and Gaussian mixture sources.

Original languageEnglish (US)
Title of host publicationProceedings of the Data Compression Conference
EditorsJames A. Storer, Martin Cohn
PublisherPubl by IEEE
Pages12-21
Number of pages10
ISBN (Print)0818656379
StatePublished - Jan 1 1994
EventProceedings of the Data Compression Conference - Snowbird, UT, USA
Duration: Mar 29 1994Mar 31 1994

Other

OtherProceedings of the Data Compression Conference
CitySnowbird, UT, USA
Period3/29/943/31/94

Fingerprint

Entropy
Annealing
Vector quantization
Free energy

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications

Cite this

Rose, K., Miller, D. J., & Gersho, A. (1994). Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle. In J. A. Storer, & M. Cohn (Eds.), Proceedings of the Data Compression Conference (pp. 12-21). Publ by IEEE.
Rose, Kenneth ; Miller, David Jonathan ; Gersho, Allen. / Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle. Proceedings of the Data Compression Conference. editor / James A. Storer ; Martin Cohn. Publ by IEEE, 1994. pp. 12-21
@inproceedings{908159a6940a4902b40179db90eb696f,
title = "Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle",
abstract = "We address the variable rate tree-structured vector quantizer design problem, wherein the rate is measured by the quantizer's entropy. For this problem, tree pruning via the Generalized Breiman-Friedman-Olshen-Stone algorithm obtains solutions which are optimal over the restricted solution space consisting of all pruned trees derivable from an initial tree. We develop a joint optimization method which is inspired by the deterministic annealing algorithm for data clustering, and which extends our previous work on tree-structured vector quantization. The method is based on the principle of minimum cross entropy, using informative priors to approximate the unstructured solution while imposing the structural constraint. As in the original deterministic annealing method, the number of distinct codevectors (and hence the tree) grows by a sequence of bifurcations in the process, which occur as solutions of a free energy minimization. Our method obtains performance gains over growing and pruning methods for variable rate quantization of Gauss-Markov and Gaussian mixture sources.",
author = "Kenneth Rose and Miller, {David Jonathan} and Allen Gersho",
year = "1994",
month = "1",
day = "1",
language = "English (US)",
isbn = "0818656379",
pages = "12--21",
editor = "Storer, {James A.} and Martin Cohn",
booktitle = "Proceedings of the Data Compression Conference",
publisher = "Publ by IEEE",

}

Rose, K, Miller, DJ & Gersho, A 1994, Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle. in JA Storer & M Cohn (eds), Proceedings of the Data Compression Conference. Publ by IEEE, pp. 12-21, Proceedings of the Data Compression Conference, Snowbird, UT, USA, 3/29/94.

Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle. / Rose, Kenneth; Miller, David Jonathan; Gersho, Allen.

Proceedings of the Data Compression Conference. ed. / James A. Storer; Martin Cohn. Publ by IEEE, 1994. p. 12-21.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle

AU - Rose, Kenneth

AU - Miller, David Jonathan

AU - Gersho, Allen

PY - 1994/1/1

Y1 - 1994/1/1

N2 - We address the variable rate tree-structured vector quantizer design problem, wherein the rate is measured by the quantizer's entropy. For this problem, tree pruning via the Generalized Breiman-Friedman-Olshen-Stone algorithm obtains solutions which are optimal over the restricted solution space consisting of all pruned trees derivable from an initial tree. We develop a joint optimization method which is inspired by the deterministic annealing algorithm for data clustering, and which extends our previous work on tree-structured vector quantization. The method is based on the principle of minimum cross entropy, using informative priors to approximate the unstructured solution while imposing the structural constraint. As in the original deterministic annealing method, the number of distinct codevectors (and hence the tree) grows by a sequence of bifurcations in the process, which occur as solutions of a free energy minimization. Our method obtains performance gains over growing and pruning methods for variable rate quantization of Gauss-Markov and Gaussian mixture sources.

AB - We address the variable rate tree-structured vector quantizer design problem, wherein the rate is measured by the quantizer's entropy. For this problem, tree pruning via the Generalized Breiman-Friedman-Olshen-Stone algorithm obtains solutions which are optimal over the restricted solution space consisting of all pruned trees derivable from an initial tree. We develop a joint optimization method which is inspired by the deterministic annealing algorithm for data clustering, and which extends our previous work on tree-structured vector quantization. The method is based on the principle of minimum cross entropy, using informative priors to approximate the unstructured solution while imposing the structural constraint. As in the original deterministic annealing method, the number of distinct codevectors (and hence the tree) grows by a sequence of bifurcations in the process, which occur as solutions of a free energy minimization. Our method obtains performance gains over growing and pruning methods for variable rate quantization of Gauss-Markov and Gaussian mixture sources.

UR - http://www.scopus.com/inward/record.url?scp=0027928553&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0027928553&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:0027928553

SN - 0818656379

SP - 12

EP - 21

BT - Proceedings of the Data Compression Conference

A2 - Storer, James A.

A2 - Cohn, Martin

PB - Publ by IEEE

ER -

Rose K, Miller DJ, Gersho A. Entropy-constrained tree-structured vector quantizer design by the minimum cross entropy principle. In Storer JA, Cohn M, editors, Proceedings of the Data Compression Conference. Publ by IEEE. 1994. p. 12-21