Abstract
A saliency-based visual attention model can simulate human eye movement. The higher saliency of the visual location, the more it attracts visual attention. While saliency map is computed from local information, ensemble information reflects spatially distributed objects. In this study, we examined whether ensemble information, in addition to saliency, influences human eye movements, and investigated the relationship between them. In the experiment, Gabor patches were placed in the left and right side of the screen with pink-noised background. The task was to count the point of Gabors (the black-white Gabor had 1 point and the colored Gabor had 2 or more points) and judge which side of the screen included just 20 points, during which their eye movements were recorded. We manipulated the difference in points between the left and right sides. In each trial, there was only one colored (the most salient) Gabor, which was located in the side with 20 points (congruent condition), or in the side with less than 20 points (incongruent condition). In the congruent condition, both saliency and ensemble information indicate the same side as target, whereas in the incongruent condition with large point difference, they indicate opposite sides as target. The results showed that participants' accuracy was consistently high in the congruent condition, whereas they increased from 57% to 94% as the point difference between sides of the screen increased in the incongruent condition. In the beginning of trials, eye movements in the incongruent condition were captured to the incorrect side, which has less than 20 points but included the colored Gabor. After a few saccades, their eyes moved to the correct side when the point difference was large. These results suggest that at first eye movements depend primarily on saliency, and ensemble information gradually affects them after a few saccades.
Meeting abstract presented at VSS 2016