Abstract
One mechanism for controlling the deployment of eye movements is attentional capture. Such capture may depend on the history of experience in the environment. Itti & Baldi (2006) have suggested that the visual system is sensitive to ‘Surprise’, that is, the deviation of the current image statistics from the distribution of prior experience. In the present study, we tested how prior experience influences eye movements to a novel object. Subjects walked in a virtual environment while their eye gaze was tracked. All subjects walked 20 laps, ∼20m long, along a oval sidewalk in a virtual environment containing 13 prominent objects placed along the path. On a critical trial, a novel object was inserted next to the path. Additionally, 2 virtual pedestrians started walking against the subject on the critical trial and remained for the duration of the experiment. For subjects in the ‘Inexperienced’ condition (n=9) this change occurred on the second lap. For those in the ‘Experienced’ condition (n=8) it occurred on the sixteenth lap. Eye movements were manually coded for four of the objects in the scene, including the novel object. Fixation durations and probabilities and saccade latencies were calculated. Across both conditions, subjects fixated both novel and non-novel objects about 2/3 of the time on any given lap for about 200–600ms. Fixation durations and saccade latencies did not differ for the novel object for different levels of experience. We found little evidence to suggest that the novel object captured gaze more than other objects in the scene independent of the subject's level of familiarity with the scene. Thus Surprise with respect to a familiar image sequence may not be a universally effective mechanism for attracting gaze. Rather, task constraints may dominate Surprise in many situations. Itti,L; Baldi, PF; Bayesian Surprise Attracts Human Attention; NIPS, 19, 547–554, 2006.