The Control of Gaze in Dynamic 1overf-Noise Displays

Christoph Rasche and Karl R Gegenfurtner

Numerous attempts have been made in the past to predict where observers fixate when viewing natural images (Itti et al 06, Tatler et al 05, Kienzle et al 06). Despite great efforts, it seems that the prediction rate for fixation vs. random patches is limited between a ROC-area value of 0.62-0.68. We used displays containing dynamic random noise to explore whether prediction performance could be higher with dynamic stimuli free of semantic meaning. We also added small local perturbations to the random noise images to see whether those could attract subjects' gaze. The noise stimuli were flickering bar codes with a spatial and temporal frequency spectrum of 1/f, mimicking the statistical properties of the real world. Small perturbation targets were generated by adding a small change to the movie, defined by luminance, color, contrast, flicker, or movement. The location of the perturbation was defined relative to the current gaze position at varying eccentricities. Several thousand saccades were recorded with an EyeLink II system from several subjects. In one series of experiments, subjects had to press a button to indicate that they had noticed the perturbation target. A classification analysis of fixated versus non-fixated movie patches showed that the prediction rate for our stimuli varied between 0.55-0.62 (ROC area), almost as high as the results obtained with static images. We did, however, find that the movie perturbations were quite effective in attracting gaze. Between 24% and 52% of the perturbations were followed by a saccade to the target stimulus. In many of these cases (up to 18%), the target stimulus went unnoticed. An analysis of the spatial distribution of fixations following target stimuli showed that center target stimuli were more effective than peripheral ones. These results indicate that it is in principle possible to steer gaze to certain image locations.

Poster in pdf format.