EFFECTS OF CONTRAST ADAPTATION ON HIGH LEVEL OBJECT IDENTIFICATION TASKS.
((J. Fiser and I. Fine)) Dept. of Brain and Cognitive Sciences and Center for Visual Science, University of Rochester, 14627-0268.
Purpose: Adaptation to the contrast of visual stimuli
has been widely demonstrated in both single cell recordings and psychophysical
experiments. Although the computational
benefits of contrast adaptation have been extensively discussed in recent years
(Heeger, 1992; Albrecht and Geisler, 1994), little is known about the role of
contrast adaptation within higher level visual tasks. Methods: Gray-scale images of everyday objects were
contrast normalized using a histogram equalization procedure based on square
wave histograms. An RSVP (Rapid Serial
Visual Presentation) paradigm was used, consisting of 192 sequences of 40
images presented at a rate of 72 msec/image.
Eight initial images were used as a 500 msec buffer. Observers had to identify whether a target
object, verbally specified before the beginning of each sequence, was
present. Sequences were either homogeneous, where all images were of
the same (either low or high) contrast, or heterogeneous,
where a single image (the target in half the sequences) was of the opposite
contrast. The mean luminance of the
monitor was 34.5 cd/m2. The luminance
range of the low contrast images varied between 29.25 and 39.75 cd/m2 and the
luminance range of the high contrast images varied between 13 and 56 cd/m2. Results: Despite a more than four-fold difference in
contrast between low and high contrast images, observers' performance did not
significantly differ between the two homogenous sequences (~79% correct for
both high and low contrast sequences).
In the heterogeneous sequences, when a single high contrast image was
embedded within a low contrast sequence, there was a highly significant
improvement in performance (~93% correct).
However, when a single low contrast image was embedded within a high
contrast sequence, performance did not deviate significantly from performance
for homogenous sequences (~74% correct).
Conclusions: These results suggest that observers
adapt to overall contrast within an approximate time frame of greater than 72
msec and less than 500 msec . This
global contrast adaptation occurs despite rapid changes (72 msec) in the local
attributes of the input (such as local contrast, orientation and spatial
frequency). Supported by NIH grant
R29-MH54770. None.