Abstract
In another presentation at this symposium, Myers and Wagner review the use of the Bayesian ideal observer as a model for human performance and define efficiency as the ratio of squares of detectability indices for the human and the Bayesian. Though this approach has proved very useful, especially in evaluation and optimization of medical imaging systems, the kinds of problems that can be analyzed in this manner are limited because the Bayesian observer needs to know the full probability laws for the data. In practice this restricts the Bayesian to fairly simple and often unrealistic tasks. Linear discriminants, on the other hand, can be calculated from knowledge of just the first and second-order statistics of the data; thus they are often analytically tractable when the Bayesian model is not. In this paper we briefly review the principles of this approach, summarize the circumstances where it is practical, and discuss a variety of psychophysical studies that show that a particular linear discriminant, which we refer to as the Hotelling observer, correlates well with human performance. Efficiency is then redefinied in terms of the Hotelling observer.
© 1992 Optical Society of America
PDF ArticleMore Like This
J. P. Rolland, H. H. Barrett, and G. W. Seeley
TuEE4 OSA Annual Meeting (FIO) 1990
Jie Yao, Harrison H. Barrett, and Jannick P. Rolland
ThAA3 OSA Annual Meeting (FIO) 1991
K. J. Myers and R. F. Wagner
FCC1 OSA Annual Meeting (FIO) 1992