A Bayesian framework for the integration of visual modules

Heinrich Bulthoff, Alan L. Yuille

Research output: Contribution to journalArticle

14 Citations (Scopus)

Abstract

The Bayesian approach to vision provides a fruitful theoretical framework both for modeling individual cues, such as stereo, shading, texture, and occlusion, and for integrating their information. In this formalism we represent the viewed scene by one, or more, surfaces using prior assumptions about the surface shapes and material properties. On theoretical grounds, the less information available to the cues (and the less accurate it is), the more important these assumptions become. This suggests that visual illusions, and biased perceptions, will arise for scenes for which the prior assumptions are not appropriate. We describe psychophysical experiments which are consistent with these ideas. Our Bayesian approach also has two important implications for coupling different visual cues. First, different cues cannot in general be treated independently and then simply combined together at the end. There are dependencies between them that have to be incorporated into the models. Second, a single generic prior assumption is not sufficient even if it does incorporate cue interactions because there are many different types of visual scenes and different models are appropriate for each. This leads to the concept of competitive priors where the visual system must choose the correct model depending on the stimulus.

Original languageEnglish
Pages (from-to)47-70
Number of pages24
JournalAttention and Performance
Volume16
Publication statusPublished - 1996 Dec 1

Fingerprint

Cues
Bayes Theorem

ASJC Scopus subject areas

  • Experimental and Cognitive Psychology
  • Neuropsychology and Physiological Psychology

Cite this

A Bayesian framework for the integration of visual modules. / Bulthoff, Heinrich; Yuille, Alan L.

In: Attention and Performance, Vol. 16, 01.12.1996, p. 47-70.

Research output: Contribution to journalArticle

@article{d963b73668d64456ae733a5b0903f795,
title = "A Bayesian framework for the integration of visual modules",
abstract = "The Bayesian approach to vision provides a fruitful theoretical framework both for modeling individual cues, such as stereo, shading, texture, and occlusion, and for integrating their information. In this formalism we represent the viewed scene by one, or more, surfaces using prior assumptions about the surface shapes and material properties. On theoretical grounds, the less information available to the cues (and the less accurate it is), the more important these assumptions become. This suggests that visual illusions, and biased perceptions, will arise for scenes for which the prior assumptions are not appropriate. We describe psychophysical experiments which are consistent with these ideas. Our Bayesian approach also has two important implications for coupling different visual cues. First, different cues cannot in general be treated independently and then simply combined together at the end. There are dependencies between them that have to be incorporated into the models. Second, a single generic prior assumption is not sufficient even if it does incorporate cue interactions because there are many different types of visual scenes and different models are appropriate for each. This leads to the concept of competitive priors where the visual system must choose the correct model depending on the stimulus.",
author = "Heinrich Bulthoff and Yuille, {Alan L.}",
year = "1996",
month = "12",
day = "1",
language = "English",
volume = "16",
pages = "47--70",
journal = "Attention and Performance",
issn = "1047-0387",
publisher = "Oxford University Press",

}

TY - JOUR

T1 - A Bayesian framework for the integration of visual modules

AU - Bulthoff, Heinrich

AU - Yuille, Alan L.

PY - 1996/12/1

Y1 - 1996/12/1

N2 - The Bayesian approach to vision provides a fruitful theoretical framework both for modeling individual cues, such as stereo, shading, texture, and occlusion, and for integrating their information. In this formalism we represent the viewed scene by one, or more, surfaces using prior assumptions about the surface shapes and material properties. On theoretical grounds, the less information available to the cues (and the less accurate it is), the more important these assumptions become. This suggests that visual illusions, and biased perceptions, will arise for scenes for which the prior assumptions are not appropriate. We describe psychophysical experiments which are consistent with these ideas. Our Bayesian approach also has two important implications for coupling different visual cues. First, different cues cannot in general be treated independently and then simply combined together at the end. There are dependencies between them that have to be incorporated into the models. Second, a single generic prior assumption is not sufficient even if it does incorporate cue interactions because there are many different types of visual scenes and different models are appropriate for each. This leads to the concept of competitive priors where the visual system must choose the correct model depending on the stimulus.

AB - The Bayesian approach to vision provides a fruitful theoretical framework both for modeling individual cues, such as stereo, shading, texture, and occlusion, and for integrating their information. In this formalism we represent the viewed scene by one, or more, surfaces using prior assumptions about the surface shapes and material properties. On theoretical grounds, the less information available to the cues (and the less accurate it is), the more important these assumptions become. This suggests that visual illusions, and biased perceptions, will arise for scenes for which the prior assumptions are not appropriate. We describe psychophysical experiments which are consistent with these ideas. Our Bayesian approach also has two important implications for coupling different visual cues. First, different cues cannot in general be treated independently and then simply combined together at the end. There are dependencies between them that have to be incorporated into the models. Second, a single generic prior assumption is not sufficient even if it does incorporate cue interactions because there are many different types of visual scenes and different models are appropriate for each. This leads to the concept of competitive priors where the visual system must choose the correct model depending on the stimulus.

UR - http://www.scopus.com/inward/record.url?scp=53349099940&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=53349099940&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:53349099940

VL - 16

SP - 47

EP - 70

JO - Attention and Performance

JF - Attention and Performance

SN - 1047-0387

ER -