Purchase this article with an account.
Jiri Lukavsky, Filip Dechterenko; The effect of scene category distinctiveness on memory performance. Journal of Vision 2015;15(12):357. doi: 10.1167/15.12.357.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous studies showed a remarkable capacity to memorize large number of scenes (Konkle et al., 2010). Same study showed exemplar effect (higher number of exemplars per category led to decreased performance). Can images of similar gist from different categories also impair the performance? Alternatively, people are trained to deal with these similarities: when they encounter a scene coming from the class of scenes of similar gist and mixed categories, they are more sensitive when encoding it. We used a set of 2048 grayscale scenes (64 categories). We evaluated the pairwise scene similarity using gist metric (Oliva & Torralba, 2001). For each image we inspected the most proximate 32 images and calculated its category distinctiveness as the proportion of the images of the same category. The distinctiveness was later averaged over categories. To each participant (N=29) we presented 320 unique scenes (5 per category) and additional 2x40 scenes for repeat detection. In the second part, the participants were making old/new judgments with 256 images (128 new, 128 old, 2+2 per category). We pooled the categories into four groups based on the distinctiveness and evaluated its effect on memory performance. We found no effect of category distinctiveness on d’. In two subsequent experiments, we found no effect in a more difficult version (12 scenes per category) or in a color version. Our experiments suggest there is no benefit in memorizing visually distinct categories (with lower gist similarity to other scene categories). The images with similar global characteristics (gist) from different categories did not serve as additional exemplars and their presence did not decrease the memory performance.
Meeting abstract presented at VSS 2015
This PDF is available to Subscribers Only