Abstract

Many previous visual signal detection and discrimination experiments have been interpreted using the concept of noise internal to the human observer. This noise is best characterized by an equivalent input noise spectral density that depends mainly on display luminance.¹ This internal noise component has been assumed to be independent of external (display) noise. Experiments are described showing that there is another component—induced internal noise—that increases in proportion to increasing external noise. The induced component dominates the other (constant or independent) component when the display noise is easily visible. The experiments were done in two ways: by comparing decisions for multiple passes through noise-limited digital image sets and also by using the same noise in both alternatives of a two- alternative forced choice decision. Induced noise estimates were the same in both cases. It was found that the induced internal noise spectral density was ~60 % of the external noise density. The results can be explained by a model quasi-ideal observer that is not able to precisely use prior information about signal parameters. The model includes a source of random signal parameter jitter about mean values followed by a maximum likelihood decision strategy.

© 1986 Optical Society of America