August 2012
Volume 12, Issue 9
Free
Vision Sciences Society Annual Meeting Abstract  |   August 2012
Understanding the intrinsic memorability of images
Author Affiliations
  • Devi Parikh
    TTI-Chicago
  • Phillip Isola
    MIT
  • Antonio Torralba
    MIT
  • Aude Oliva
    MIT
Journal of Vision August 2012, Vol.12, 1082. doi:10.1167/12.9.1082
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Devi Parikh, Phillip Isola, Antonio Torralba, Aude Oliva; Understanding the intrinsic memorability of images. Journal of Vision 2012;12(9):1082. doi: 10.1167/12.9.1082.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Artists, advertisers, and photographers are routinely presented with the task of creating an image that a viewer will remember. While it may seem like image memorability is purely subjective, recent work shows that it is not an inexplicable phenomenon: variation in the memorability of images is consistent across observers. Some images are intrinsically more memorable than others, independent of observers’ contexts and biases. In this work, we are interested in understanding what characteristics of images make them more memorable than others.

We used the publicly available memorability dataset of Isola et al., and augmented the object and scene annotations with interpretable spatial, content, and aesthetic image properties. We used a feature selection scheme with desirable explaining-away properties to determine a core set of attributes that concisely characterizes the memorability of an image. In particular, we compared two greedy feature selection methods: 1) selecting the set of features that maximizes an approximation of the mutual information between the feature set and memorability, and 2) selecting the set of features that maximizes the performance of a memorability predictor. Our selected features are compact, yet effective at predicting memorability. They perform significantly better than randomly selected sets of features with comparable complexity.

Since our features were interpretable attributes, our selections allow for a simple and understandable explanation of memorability. We find that images of active, enclosed spaces containing people are memorable, while images of peaceful scenes with no "story" are not. Contrary to popular belief, "unusual" and aesthetically pleasing scenes do not tend to be highly memorable. We also predicted these characteristics of images automatically from low-level image-features, resulting in a fully automatic approach to predicting memorability. This work represents one of the first attempts at understanding intrinsic image memorability, and lies at an emerging interface between human cognition and computer vision.

Meeting abstract presented at VSS 2012

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×