I've been trying to determine what is the minimum presentation time of a
visual stimulus that will allow a human observer to extract *some*
information. Of course, this is going to be very task- and
experience-dependent, but for recognizing simple forms or making simple
discriminations, what is generally accepted as being a valid lower limit on
exposure time?
>From what I've been able to determine, 20 msec exposure, when followed by a
pattern mask, is pretty much the shortest feasible viewing time. This comes
from:
* Sperling, in his visual search work in the early 1970's, used a technique
of flashing arrays of letters and asking observers to detect the presence
of a numeral in one of the arrays [G. Sperling & M.J. Melchner, "Visual
search, visual attention, and the attention operating characteristic",
_Attention and Performance VII_, ed. Jean Requin, 1978]. Each array in a
sequence contained a set number of letters, from two to twenty-five, and
was flashed for a fraction of a millisecond, with a uniform (grey?) field
displayed between presentations of arrays for a (constant) period varying
from 10 msec to 320 msec. Because the arrays were presented sequentially
back to back, each new stimulus acted as a pattern mask for the previous
one. In his experiments, observers seemed to have a non-zero probability
of being able to detect a target with 20 msec presentation times in 2-item
arrays.
* Biederman, in work in the late 1970's, used a forced-choice scene
recognition test with presentation times as low as 20 msec (e.g.,
[Biederman, Rabinowitz, Glass and Webb, "On the information extracted from
a glance at a scene", J. Exp. Psych., 103(3):597-600, 1974].) One of the
experiments involved presenting pictures of common scenes (e.g., a
``shopping plaza'') to subjects for 20 to 4000 msec, followed by a pattern
mask for 300 msec, and then followed by a discrimination test in which the
subject had to select between two verbal labels (words or phrases)
describing the scene. When the scene labels were dissimilar (e.g.,
``shopping plaza'' versus ``kitchen''), subjects averaged approximately
65% correct responses at 20 msec and over 70% correct at 50 msec.
Performance, however, was worse when the labels were similar (e.g.,
``shopping plaza'' versus ``busy road and stores''); subjects averaged 55%
correct responses at 20 msec and 65% correct at 50 msec.
But 20 msec seems awfully short to me, so I wanted to check with others and
ask if this seems reasonable.
--
Mike Hucka (michael.hucka at umich.edu)
University of Michigan, Ann Arbor, Mich.
