Abstract

The Bayesian or ideal observer is a construct from signal detection theory. By definition, the ideal observer uses all information present in an image to make optimal decisions regarding the underlying scene. By comparing the calculated performance of the ideal observer to the measured performance of a human observer, we can determine how well a displayed image is utilized by the human observer. Human performance has been shown to be highly efficient for certain, detection and discrimination tasks. Degradations in human performance due to correlated noise or uncertainty in the task have been explained by using this Bayesian framework. This talk will describe the ideal-observer decision function for various tasks and methods for calculating the ideal-observer SNR. Existing data comparing the ideal and human observers for a variety of visual tasks will be reviewed.

© 1992 Optical Society of America