How to Select Both Objects and Locations

Weidong Shi & Kyle R. Cave

A poster presented at the annual Psychonomic Society meeting, Dallas, Nov 19 - 22, 1998.

Summary

We describe a new neural network model of visual attention that can facilitate object recognition when multiple visual objects are presented simultaneously. Through a combination of feedforward and feedback connections, the model can eliminate the interference among multiple objects by selecting the locations that match the currently active higher-level object representation. Thus, the model offers an alternative way to account for object selection by assuming that visual selection is primarily location based but guided by higher-level object information and viewer's expectations.

Goals of this study

• To demonstrate that object-based selection can be accomplished within a system that ultimately selects by location.
  
• To extend the FeatureGate model of attention from feature driven location based selection to object driven location based selection (Cave, in press).
  
• To show that visual selection can take the form of inhibition of distractor locations (Cepeda, Cave, Bichot & Kim, 1998).

Background

Components of the Model

How Search is Simulated in the Model

Results from Computer Simulations

Conclusions

• When a visual display contains multiple objects, detection of a target can be facilitated by inhibition of distractor locations.
  
• Such inhibition can be spatially based but guided by top down expectations.

References

Thanks to Randolph Blake, Narcisse Bichot, Carolyn Backer Cave, Thomas Palmeri and Ken Sobel for helpful suggestions.

For more information, contact:

Kyle R. Cave
University of Massachusetts
Department of Psychology
Amherst, MA 01003
U.S.A.

phone: 413-545-2787

email: kcave@psych.umass.edu