Finding diagnostic features in noise

Lizann Bonnar, Philippe G Schyns

Research output: Contribution to conferenceAbstractpeer-review

Abstract

Improvements in signal discrimination due to perceptual learning have been described as either a reduction in internal noise or an enhancement of discriminant signal features (Dosher and Lu, 1999Vision Research393197 ^ 3221; Gold et al, 1999 Nature402176 ^ 178). Following the proposal of signal enhancement, those features enabling fine discriminations between incoming signals are diagnostic to resolve the task. We are here interested in the relationship between the diagnostic status of a feature and the perceptual learning mechanism that enables its extraction. We first derived, using the `bubbles' technique of Gosselin and Schyns (2001Vision Research412261 ^ 2271), the diagnostic regions used by human observers to perform a face-identity task. In a perceptual learning paradigm, using the same faces, new observers performed a 10AFCface-identity task in which we varied the information available to perform the task by applying the diagnostic mask to the face. To distinguish between the potential candidates behind improved performance we varied the contrast energy of the signal and the level of external noise added to the stimulus. Performance thresholds of the human observer were then compared with that of an ideal observer that uses all of the available information in the stimulus.
Original languageEnglish
Pages18-18
Number of pages1
Publication statusPublished - 25 Sept 2002
Event25th European Conference on Visual Perception - Glasgow Royal Concert Hall, Glasgow, United Kingdom
Duration: 25 Sept 200229 Sept 2002
http://ecvp.psy.gla.ac.uk

Conference

Conference25th European Conference on Visual Perception
Country/TerritoryUnited Kingdom
CityGlasgow
Period25/09/0229/09/02
Internet address

Keywords

  • signal enhancement
  • external noise
  • face recognition

Fingerprint

Dive into the research topics of 'Finding diagnostic features in noise'. Together they form a unique fingerprint.

Cite this