A moving selection window allows one to exceed the normal temporal limits on binding

from Cavanagh, P. & Holcombe, A. O., Distinguishing pre-selection from post-selection processing limits using a moving window of selection , 2005 meeting of the Vision Sciences Society and 2008 Mobile computation: Spatiotemporal integration of the properties of objects in motion. Cavanagh, P., Holcombe, A.O., & Chou, W-L., Journal of Vision. 8(12):1-23, doi:10.1167/8.12.1



While gazing directly at the blue spot in the displays below, try to determine whether the green patches are paired with rightward-tilted bars or with leftward-tilted bars. In the actual experimental display, which will resemble this display to varying degrees depending on your browser etc., observers could pair the features at much faster rates with the display on the right.

In both displays, a red patch alternates with green patch adjacent to a set of bars alternating between right- and leftward tilted. Holcombe & Cavanagh (2001) found that observers were unable to determine the feature pairing when the stimuli alternated at rates faster than about 6 per second. The present, newer displays show that this temporal limit is imposed at a very late stage in visual processing.

For the display on the left, observers are asked to report the feature pairing outlined by the blue circle. Most people can only perform reliably when the presentation rate is slower than 6 stimuli per second. Is this because the representation of the features or their temporal registration is degraded within the visual system? The display on the right, which is identical to the left display but for the movement of the circle, shows that this is not the case.

The blue circle serves as a moving guide for attention, always encircling the green half of the alternation. The circle likely leaves the local analysis of the features early in the visual system largely unchanged. However, the effect on more global processing stages is dramatic. At these more cognitive stages, the rapid alternation of features is transformed into a constant representation, and the task becomes much easier. Note that this could only happen if the representation of the features in the "difficult" case was intact locally, but simply could not be reported due to difficulties at later stages of processing.

back to publications