Mining Videos for Features that Drive Attention

Abstract

Certain features of a video capture human attention and this can be measured by recording eye movements of a viewer. Using this technique combined with extraction of various types of features from video frames, one can begin to understand what features of a video may drive attention. In this chapter we define and assess different types of feature channels that can be computed from video frames, and compare the output of these channels to human eye movements. This provides us with a measure of how well a particular feature of a video can drive attention. We then examine several types of channel combinations and learn a set of weightings of features that can best explain human eye movements. A linear combination of features with high weighting on motion and color channels was most predictive of eye movements on a public dataset.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Apr 01, 2015
Accession Number
AD1002651

Entities

People

  • Farhan Baluch
  • Laurent Itti

Organizations

  • University of Southern California

Tags

Communities of Interest

  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Artificial Intelligence
  • Cognition
  • Computer Programs
  • Computer Science
  • Computer Vision
  • Computers
  • Data Mining
  • Electronic Mail
  • Eye Movements
  • Genetic Algorithms
  • Military Research
  • Object Recognition
  • Orientation (Direction)
  • Video
  • Video Frames
  • Video Games

Fields of Study

  • Computer science

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Computer Vision.