Abstract
Sensory memory refers to a modality-specific neural representation of sensory stimuli in the human brain. Iconic memory - very short term visual memory - historically was measured in noiseless, high-contrast displays of multiple items using partial report cuing methods (Sperling, Psycholog. Mono.,1960). Here, we derive an observer model for iconic memory at a range of contrasts in order to characterize the decaying perceptual representation. We measure the duration of iconic memory for basic visual stimuli at a range of contrasts and with and without external noise. The visual displays consisted of an annular arrangement of 8 Gabor patches, all eight displayed either with or without external noise and at one of three contrast levels. Each Gabor could appear in one of four orientations, chosen independently. Very short-term visual or iconic memory was measured through a partial report procedure in which one Gabor was cued for report of orientation at one of 8 delays between 0 (simultaneous cue) and 3.2 sec. The three contrasts, varying across trials, were chosen separately for no-noise and high noise displays to yield approximately 60, 75, and 90 percent correct based on individual observer pretests. Four observers were tested. For all 6 (2 noise × 3 contrast) conditions, iconic memory decayed exponentially over the maximum test interval of 3.2 seconds with the same or nearly the same decay constants of about 0.3 sec. This rapid decay in multi-element displays is consistent with early estimates of the persistence of iconic memory. The systematic relationship between noise and contrast conditions was well accounted for by a divisive gain-control formulation of the Perceptual Template Model (Lu & Dosher, Vis. Res. 1998) in which an initial encoded representation of the stimulus decays exponentially, with constant, late noise.