Correspondence computations in visual cognition
Embargo until
Date
2014-03-19
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
15 behavioral experiments were conducted to investigate the role of object
correspondence computations in visual cognition. Correspondence computations refer,
here, to algorithms that identify relationships between objects in temporally separate
encounters.
In Experiment 1-5, I hypothesized that tracking failures occur because of
correspondence failures during close encounters of targets and nontargets. To test this
idea, I provided observers with different surface feature information to nontargets
whenever they approached within 4° of a target (Experiment 1). This manipulation
significantly improved performance by alleviating correspondence challenges. Two
control experiments showed that this color change benefit is not merely due to target
recovery (Experiment 2 and 4). A follow-up experiment measured the distance at which
objects correspondence becomes challenging (Experiment 3). And an additional
experiment demonstrated that the overall frequency of target-nontarget close encounters
predict human performance (Experiment 5).
Experiment 6-10 explored the role of object correspondence in the context of
spatial working memory. Experiment 6 supplied evidence of object correspondences in a
typical spatial working memory task through a trial specific analysis. In addition, a model
that implements correspondence algorithms successfully predicted human performance
without assuming any independent memory-related limits. Experiments 7 and 8
employed a preview display that indirectly provided information about memory location
to be tested. This manipulation improved SWM performance dramatically (e.g.
performance with 8 objects were comparable to 2 objects). A control experiment showed
that the improved performance is not due to mere reactivation of memory representations (Experiment 9). Additional experiment showed that object colors do not support
correspondence computations in this context.
Experiment 11-15 employed integral features to prevent correspondence failures
in a visual working memory task. I reasoned that integral features can be used to solve
correspondence problems by preventing confusions between objects. Experiment 11 and
12 independently identified integral features using perceptual sorting experiments. When
these features were used in change judgment tasks, working memory with two objects
produced performance as precise as with one (Experiments 13-15).
Taken together, these results suggest that object correspondence play a crucial role in the
constraints typically observed in visual cognition.
Description
Keywords
Visual cognition, visual working memory, multiple object tracking, correspondence problems, vision