| |
Abstract:
Abstract: Observers can easily find a simple pattern of dots
amidst random dot noise, particularly if the test pattern appears
at a known location (Uttal, 1975). It is much harder to find the
target pattern when it is presented at an unknown location, because
the observer must monitor many potential locations to find it. In a
search task, we measured accuracy for finding an oblique target
string of five aligned dots presented at an unknown location, as a
function of the number of distractors. We used two kinds of
distractors -- oriented dot pairs, and randomly-placed dots.
Observers judged which of two 200 ms intervals contained the target
string. The accuracy for finding the target string among oriented
distractors of like orientation was greatly degraded compared to
the accuracy for finding the target among randomly-placed
distractors. Orthogonally-oriented distractors had almost the same
effect as randomly-placed distractors. These results suggest that
search for this simple pattern is based on competing responses in
oriented neural units in early vision. We successfully modeled
these results with an array of multi-scaled oriented units
optimally tuned for the target string. To compare the computational
results with the search data, the model "chose" the interval with
the largest "neural" response. The model also predicts that search
should be scale-invariant, in agreement with our results showing
nearly constant performance for distances ranging from 0.5 - 2
meters.
|
