Scale mixtures of Graussians and the statistics of natural images

Martin J. Wainwright, Eero Simoncelli

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

Abstract

The statistics of photographic images, when represented using multiscale (wavelet) bases, exhibit two striking types of non-Gaussian behavior. First, the marginal densities of the coefficients have extended heavy tails. Second, the joint densities exhibit variance dependencies not captured by second-order models. We examine properties of the class of Gaussian scale mixtures, and show that these densities can accurately characterize both the marginal and joint distributions of natural image wavelet coefficients. This class of model suggests a Markov structure, in which wavelet coefficients are linked by hidden scaling variables corresponding to local image structure. We derive an estimator for these hidden variables, and show that a nonlinear "normalization" procedure can be used to Gaussianize the coefficients.

Original languageEnglish (US)
Title of host publicationAdvances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999
PublisherNeural information processing systems foundation
Pages855-861
Number of pages7
ISBN (Print)0262194503, 9780262194501
StatePublished - 2000
Event13th Annual Neural Information Processing Systems Conference, NIPS 1999 - Denver, CO, United States
Duration: Nov 29 1999Dec 4 1999

Other

Other13th Annual Neural Information Processing Systems Conference, NIPS 1999
CountryUnited States
CityDenver, CO
Period11/29/9912/4/99

Fingerprint

Statistics

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Information Systems
  • Signal Processing

Cite this

Wainwright, M. J., & Simoncelli, E. (2000). Scale mixtures of Graussians and the statistics of natural images. In Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999 (pp. 855-861). Neural information processing systems foundation.

Scale mixtures of Graussians and the statistics of natural images. / Wainwright, Martin J.; Simoncelli, Eero.

Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999. Neural information processing systems foundation, 2000. p. 855-861.

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

Wainwright, MJ & Simoncelli, E 2000, Scale mixtures of Graussians and the statistics of natural images. in Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999. Neural information processing systems foundation, pp. 855-861, 13th Annual Neural Information Processing Systems Conference, NIPS 1999, Denver, CO, United States, 11/29/99.
Wainwright MJ, Simoncelli E. Scale mixtures of Graussians and the statistics of natural images. In Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999. Neural information processing systems foundation. 2000. p. 855-861
Wainwright, Martin J. ; Simoncelli, Eero. / Scale mixtures of Graussians and the statistics of natural images. Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999. Neural information processing systems foundation, 2000. pp. 855-861
@inproceedings{ca3b4dc09d1f4a62ae9889390c4c6d95,
title = "Scale mixtures of Graussians and the statistics of natural images",
abstract = "The statistics of photographic images, when represented using multiscale (wavelet) bases, exhibit two striking types of non-Gaussian behavior. First, the marginal densities of the coefficients have extended heavy tails. Second, the joint densities exhibit variance dependencies not captured by second-order models. We examine properties of the class of Gaussian scale mixtures, and show that these densities can accurately characterize both the marginal and joint distributions of natural image wavelet coefficients. This class of model suggests a Markov structure, in which wavelet coefficients are linked by hidden scaling variables corresponding to local image structure. We derive an estimator for these hidden variables, and show that a nonlinear {"}normalization{"} procedure can be used to Gaussianize the coefficients.",
author = "Wainwright, {Martin J.} and Eero Simoncelli",
year = "2000",
language = "English (US)",
isbn = "0262194503",
pages = "855--861",
booktitle = "Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999",
publisher = "Neural information processing systems foundation",

}

TY - GEN

T1 - Scale mixtures of Graussians and the statistics of natural images

AU - Wainwright, Martin J.

AU - Simoncelli, Eero

PY - 2000

Y1 - 2000

N2 - The statistics of photographic images, when represented using multiscale (wavelet) bases, exhibit two striking types of non-Gaussian behavior. First, the marginal densities of the coefficients have extended heavy tails. Second, the joint densities exhibit variance dependencies not captured by second-order models. We examine properties of the class of Gaussian scale mixtures, and show that these densities can accurately characterize both the marginal and joint distributions of natural image wavelet coefficients. This class of model suggests a Markov structure, in which wavelet coefficients are linked by hidden scaling variables corresponding to local image structure. We derive an estimator for these hidden variables, and show that a nonlinear "normalization" procedure can be used to Gaussianize the coefficients.

AB - The statistics of photographic images, when represented using multiscale (wavelet) bases, exhibit two striking types of non-Gaussian behavior. First, the marginal densities of the coefficients have extended heavy tails. Second, the joint densities exhibit variance dependencies not captured by second-order models. We examine properties of the class of Gaussian scale mixtures, and show that these densities can accurately characterize both the marginal and joint distributions of natural image wavelet coefficients. This class of model suggests a Markov structure, in which wavelet coefficients are linked by hidden scaling variables corresponding to local image structure. We derive an estimator for these hidden variables, and show that a nonlinear "normalization" procedure can be used to Gaussianize the coefficients.

UR - http://www.scopus.com/inward/record.url?scp=84898965689&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84898965689&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84898965689

SN - 0262194503

SN - 9780262194501

SP - 855

EP - 861

BT - Advances in Neural Information Processing Systems 12 - Proceedings of the 1999 Conference, NIPS 1999

PB - Neural information processing systems foundation

ER -