de.mpg.escidoc.pubman.appbase.FacesBean
Deutsch
 
Hilfe Wegweiser Impressum Kontakt Einloggen
  DetailsucheBrowse

Datensatz

DATENSATZ AKTIONENEXPORT

Freigegeben

Vortrag

Sensitivity to local higher-order correlations in natural images

MPG-Autoren
http://pubman.mpdl.mpg.de/cone/persons/resource/persons84519

Gerhard,  H
Research Group Computational Vision and Neuroscience, Max Planck Institute for Biological Cybernetics, Max Planck Society;

http://pubman.mpdl.mpg.de/cone/persons/resource/persons84314

Wichmann,  F
Department Empirical Inference, Max Planck Institute for Biological Cybernetics, Max Planck Society;

http://pubman.mpdl.mpg.de/cone/persons/resource/persons83805

Bethge,  M
Research Group Computational Vision and Neuroscience, Max Planck Institute for Biological Cybernetics, Max Planck Society;

Externe Ressourcen
Es sind keine Externen Ressourcen verfügbar
Volltexte (frei zugänglich)
Es sind keine frei zugänglichen Volltexte verfügbar
Ergänzendes Material (frei zugänglich)
Es sind keine frei zugänglichen Ergänzenden Materialien verfügbar
Zitation

Gerhard, H., Wichmann, F., & Bethge, M. (2011). Sensitivity to local higher-order correlations in natural images. Talk presented at 34th European Conference on Visual Perception. Toulouse, France.


Zitierlink: http://hdl.handle.net/11858/00-001M-0000-0013-BAAA-7
Zusammenfassung
We measured perceptual sensitivity to higher-order correlational structure of natural images using a new paradigm, with which we also evaluated the efficacy of several successful natural image models that reproduce neural response properties of the visual system. To measure sensitivity to local correlations in natural images, stimuli were square textures of tightly tiled small image patches originating from either: (i) natural scene photographs or (ii) a model. In a trial, observers viewed both texture types and had to select the one made of natural image patches. In a series of experiments with twenty-two subjects, we tested 7 models, varying patch size from 3×3 to 8×8 pixels. Results indicate high sensitivity to local higher-order correlations in natural images: no current model fools the human eye for patches 5×5 pixels or larger, and only the model with the highest likelihood brings performance near chance when patches are 4×4 pixels or smaller. Remarkably, the ordering of the psychophysical matched the models' ordering in likelihood of capturing natural image regularities. The subjects' performance on binarzed textures approached ideal observer efficiency, where the ideal observer has perfect knowledge of the natural image distribution. In four control experiments, we determined the knowledge observers use to detect higher-order correlations.