Abstract
Extracting a target object from a cluttered background is a complex task that is a significant problem for theories of object recognition. For most theories to function properly, the human visual system needs to be capable of highly accurate segmentation that takes place prior to attempting to recognize the target object. If this is true, we would expect relatively little effect of adding irrelevant background objects to a shape discrimination task, as long as the target object is not obscured.
In our study, the background consisted of objects placed behind a target object such that the target always partially occluded each of the background objects. Observers were asked to look at the target, in the center of three background objects; the image was then masked. A second image was then shown that could have either the same or a different target shape, and either the same background objects or different ones. This image was masked, and observers were then asked to respond “same” or “different” to the target object while ignoring the background objects.
The pattern of results obtained demonstrates that segmentation was not perfect and that the background interfered with accurate task performance. Comparing performance with and without background objects, we observed a decrease in accuracy of approximately 10% when background objects were added. While this decrease was significant, performance was still well above chance, which suggests that the visual system is fairly efficient at segmentation when the background is present. There was also an unexpected interaction effect: accuracy in correctly responding “different” to a shape change in the target object increased by approximately 20% when the background changed as well. Our results suggest that the observers were unable to completely filter out the background; for at least some trials, they were seduced into basing their response on the background rather than the target object.