Abstract
We investigated differences in the propensity for human and monkey eye movements to be guided by bottom-up stimulus properties or cognitive factors. We recorded eye movements from 4 humans and 4 monkeys freely viewing 50 natural video clips (approximately 20,000 frames). To quantify species differences in attentional allocation we used a computational model of visual surprise (Itti, Baldi, 2005). We measured the amount of surprise at the endpoints of saccadic eye movements and compared it to the amount of surprise at randomly shuffled eye positions. We found that humans and monkeys are attracted to surprising locations significantly above chance (p[[lt]]10−10) but with no significant species difference (p=0.5); however, interobserver agreement is significantly higher in humans than monkeys. When multiple monkeys simultaneously agreed on a common gaze target the location had significantly increased surprise (p=.0001), while humans showed slightly decreased surprise (p=0.6) for agreed gaze locations. This indicates that monkeys agree on gaze targets for bottom-up and humans for cognitive factors. This is further supported by the observation that those strong attractors of gaze corresponded to different locations for humans and monkeys. We found that over the course of a clip monkeys remain consistently surprise driven, while humans showed significant differences over time (p[[lt]]10−10). We observed a similar trend in interobserver agreement. We conclude that humans and monkeys are in general equally driven by surprise, but humans tend to agree on locations containing scene specific semantic information while monkeys agree on locations containing visually surprising information. The time course of surprise and interobserver predictability for each species implies that top-down influences play a greater role in modulating the influence of bottom-up attention in humans than monkeys. This study indicates that monkeys serve as a good model for bottom-up attention in humans due to the limited effect of top-down influences.
National Science Foundation