Purchase this article with an account.
Laura W. Renninger, Preeti Verghese, James Coughlan; Modeling eye movements in a shape discrimination task. Journal of Vision 2005;5(8):921. doi: 10.1167/5.8.921.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We investigated the pattern and sequence of eye movements that humans produce when they study a novel object. We have presented sequential information maximization (SIM) as a viable eye movement planning strategy (Renninger & Malik, VSS 2004; Renninger, Coughlan, Verghese & Malik, NIPS 2004). SIM selects fixation locations that will maximize information about the global stimulus. This strategy causes implicit inhibition-of-return (IOR) behavior, because fixation will not return to a location unless it offers new information. We evaluate the model with a shape discrimination task. The stimuli are novel, high contrast silhouettes that subtend 12.5 deg to force observers to make eye movements. A discrimination task ensures that the figures are studied carefully. Observers fixate a marker while the silhouette is presented peripherally for 300ms. The marker disappears to cue the observer to study the shape for 1.2 seconds while their eye movements are recorded. Following the study interval, the shape just studied must be identified from two possible choices. The distractor shape is highly similar to the target, but differs slightly in its contour. We compared the eye movements of three observers to the predictions of the SIM model and found a good alignment. As a control, we also compared the results to predictions from a saliency model (Itti & Koch, 2001). This model uses center-surround contrast to score salient hotspots on or around the figure. An explicit IOR mechanism creates a sequence of eye movements to points in order of decreasing salience. The saliency model also shows a good alignment with the human data, but neither model exactly captures their behavior. Superimposing the eye movement traces on spatial prediction maps from the two models clearly indicates that including 1) the eccentric fall-off of acuity and 2) the rate at which observers incorporate the information from a fixation will greatly improve the alignment.
This PDF is available to Subscribers Only