Reliability of Human Visual Signal Detection in the Presence of Noise

This report is concerned with the reliability of human detection of certain classes of signals of simple shape embedded in low-pass filtered white gaussian noise recorded on paper. Results are compared with performance of an optimum matched filter detector which operated on the same set of data. The signal-to-noise ratio obtained at the output of such a filter can be used to predict the reliability of detection and the false alarm rate in subjects´ performance. For signal-to-noise ratio greater than one, subjects make fewer errors (i.e., misses plus false alarms) than the optimum filter, and the difference is a monotonically decreasing function of the signal-to-noise ratio. Factors concerned with these two observations are discussed.