An image's memorability is determined by the extent to which its features are distinct from those of other images. However, a mechanistic understanding of the effect of distinctiveness on visual memory has been hampered by difficulties in quantifying exactly the features defining complex, real-world scenes. In this project, we will quantify a scene's visual features and assess its distinctiveness or similarity to other scenes using deep neural network models. In order to target scenes with specific features and distinctiveness characteristics with unprecedented precision, we will additionally generate synthetic, realistic-looking scenes using generative adversarial neural networks. These computational approaches will allow us to test quantitative models addressing long-standing questions in cognitive psychology: what makes an image a typical exemplar, what makes an image distinct, and what makes an image memorable?
DFG Programme
Research Grants
International Connection
Czech Republic
