Abstract
At VSS2018, we demonstrated that of the various scene categories (aerials, cityscapes, indoors, landscapes, people), observers rated landscapes as most interesting and indoor scenes as least interesting. Here we ask: are there eye-movement proxies of viewer interest, what makes an individual scene interesting to a viewer, and can information present in the scene predict its interestingness? We studied eye scan patterns while 44 observers free viewed scenes and found that the average duration of fixations best matched subjective interest ratings, more so than number of fixations/saccades, average saccade duration or viewing duration. New stimuli grab our attention. Does presenting a scene from a new, altogether different category spike viewer interest? For this, 68 observers viewed ten scenes of people, then one scene from a non-people category, and the sequence was repeated three additional times (44 scenes total). Interest ratings of the intermittently shown non-people scenes were significantly higher than those of the more frequently shown scenes of people. In a separate experiment, scenes of people were replaced by landscapes with a similar 10:1 landscape to non-landscape ratio. Interest ratings (63 observers) of the intermittently shown non-landscapes were significantly lower than those of the more frequently shown landscapes. In summary, novelty of image category does not spike viewer interest, but rather the actual image category does. Note that viewing durations were not commensurate with interest ratings. Are more complex images more interesting? We presented two scenes side by side (40 pairs) and had 40 observers free view the scenes while they were displayed for a fixed duration. Each pair was carefully chosen so that the two images differed in Jansen-Shannon complexity. More time was spent viewing the more complex image of the pair on ~70% of trials. Further analysis is ongoing.