Abstract

Many psychophysical studies of the ability of a human observer to detect an exactly specified signal in a uniform background with Poisson noise have been reported. Most of their results can be adequately explained by an ideal-observer model, which, for this problem, reduces to a simple, linear matched filter. If, however, the background is spatially in homogeneous ("lumpy"), then the ideal observer becomes nonlinear, and its performance is usually very difficult to calculate. Because inhomogeneous backgrounds are very important in many practical applications, we have investigated the performance of human observers on this problem as a function of the amount of background in homogeneity, the blur of the imaging system, and the amount of Poisson noise in the image. We find that all of our results are well described by the assumption that the human observer behaves as the optimum linear observer (Fisher and Hotelling). A simpler linear model, the so-called nonprewhitening matched filter, predicts different behavior.

© 1990 Optical Society of America