Eyetracking data

We analyzed the eye-tracking dataset collected in [26]. The eye-tracking data was obtained for 82 observers from different age groups. All observers had normal or corrected-to-normal vision. Participants were assigned to 4 different groups: four-six years, six-eight years, eight-ten years, and adults (mean age, 29 years). We use 4 years, 6 years, 8 years, and adults to refer these groups in order. The study was conducted in conformity with the Code of Ethics of the World Medical Association (Declaration of Helsinki) and approved by the Ethics Committee of the University of Paris Descartes. All participants or their parents in case of children gave written informed consent prior to participation.

The age group assignment was made based on the findings of previous developmental studies suggesting that eye-movement control changes rapidly at the beginning i.e. during childhood, and later more slowly ([37] [38] [39] [23] [24] [19]). These previous findings were also replicated in our analysis results, where we found that the explorativeness, agreement score, and center bias tendency changes significantly during 4 to 10 years of age and gets mature after the age of 8-10 years. These results motivated us to investigate the children age groups in relatively smaller intervals to quantify the significant changes in scene viewing behavior and subsequently reflecting these changes in the age-adapted saliency model.

The experiment was conducted on images of 1024 × 764 pixels. The images were taken from children’s books and movies, and characterized to have eventful backgrounds. Our choice of images might be less interesting for adults than children. However, we decided to use these images since they were suitable for children and they can also be used in eye-gaze study in adults [40], [41], [42]. Paintings [40] and artificial stimulus [41, 42] have been used in these studies to revel the eye movement behavior in adults. These supporting studies provided us an evidence of the suitability of the image type used in the proposed study.

Further, to avoide stimulus related bias and to maintain the motivation of our participants, a segment recognition test was performed during the experiment. In which after the presentation of the image (10s) an image segment was presented at the center of the screen during 5 seconds; in 50% of the cases the image segment was valid. Participants had to determine if the segment was part of the previous scene or not by pressing a button. The results of the task performance reported in [26] suggested the high level of engagement for the selected stimuli for all age groups including adult observers, which also confirms the age appropriateness of our selected stimuli.

The remote eye-tracking system EyeLink 1000 with a sampling rate of 500 Hz was used to measure eye gaze, and provided us with the raw data that was sampled to obtain fixations and saccades. The spatial resolution of eye tracker was below 0.01°, and spatial accuracy more than 0.5°. The random fixations and noise were discarded by processing the raw data by fixation detection algorithm supplied by SR research (EyeLink).

During eye tracking experiment pictures were presented at a distance of 60 cm from a screen at a resolution of 1024 × 728, a five point calibration and validation was performed before starting the viewing task and subjects were asked to explore the scene which was presented for 10 seconds. After this time, the scene was replaced by an image segment and participants had to determine if the segment was part of previous scene or not.

For each image fixation landings of all observers were used to generate two maps for different age groups: a human fixation map and a human saliency map. The human fixation map was created as a binary representation of fixation locations, and the human saliency map was obtained by convolving a Gaussian filter across the fixation locations, as in [4]. The visualizations of human fixation and human saliency maps are shown in Fig 2. These maps were used to analyze eye-movement behavior.

The process of generating human fixation map and human saliency maps of an image for age groups. OB_n stands for the n^th observer of an age group. The images used in this figure are similar but not identical to the original image, and is therefore for illustrative purposes only.