STATISTICALLY DEFINED BACKGROUND - PERFORMANCE OF A MODIFIED NONPREWHITENING OBSERVER MODEL

Research on human-observer performance for noise-limited tasks (such as those found in medical imaging) has recently progressed to investigations in which some signal or image parameters are statistically defined. In these cases the ideal-observer procedure is usually nonlinear, and analysis is mathematically intractable. Two suboptimal but linear observer models have been proposed for mathematical convenience. The Hotelling observer is the optimal linear model and has been found to give a good fit to most human results. The nonprewhitening (NPW) matched filter also has been useful for explanation of some human results. Rolland and Barrett [J. Opt. Soc. Am. A 9, 649 (1992)] recently reported human results for detection of signals in white noise superimposed on statistically defined (lumpy) backgrounds in experiments that simulated nuclear medicine imaging systems. They found that the Hotelling model gave a good fit, whereas the simple NPW matched filter gave a poor fit. It is shown that the NPW model can also fit their data if a spatial frequency filter of a shape similar to the human contrast-sensitivity function is added to the NPW observer model. The best fit is achieved by use of an eye-filter model E(f) = f1.3 exp(-cf2), with c selected to yield a peak at 4 cycles/deg.