Abstract
Some images we see stick in mind, while others fade. Recent studies of visual memory have found remarkable levels of consistency for this inter-item variability across observers (e.g., Isola 2011), suggesting that memorability can be considered an intrinsic image property. However, the visual features underlying memorability are not yet well understood. Investigating the relation between image memorability and inter-item variability in other visual tasks can provide more insight. Here, we asked whether an image that is easier to process and categorize is also more memorable. We used a rapid-scene categorization task and assessed whether there are consistent differences in difficulty between images in this task (defined as "categorizability") and whether they correlate with memorability. We selected 14 scene categories and 44 images per category from a set previously quantified on memorability (Bylinskii 2015). Per trial, participants saw an image for a duration of 32 ms, followed by a mask of 80 ms. Next, a category label appeared on screen and the task was to indicate whether the label matched the image. For each participant, a random half of the scenes was presented as signal trials (i.e., label matches image), the other half as no-signal trials. For signal trials, we collected on average 79 responses per image. An image's categorizability score was calculated as the proportion of correct responses on signal trials. The average categorizability score per category varied between .55 and .89. Thus, given the task context, some categories were considerably easier than others. For most categories, consistency scores were high (mean split-half Spearman's rho up to .90), suggesting that categorizability is an intrinsic image property too. However, the predicted positive correlation between categorizability and memorability was not observed. This suggests that the ease with which an image can be categorized relies on features distinct from those involved in memorability.
Meeting abstract presented at VSS 2017