Exploring New Dimensions of Archives: Finding Audiovisually Similar Programmes with the Help of Neural Networks

Abstract

Searching for programs with audiovisually similar keyframes offers a completely new way of finding related content. It enables researchers to analyse the re-use and spread of imagery. We have implemented a ‘Similarity’ tool in the Media Suite, a Research Environment for analysing content from audiovisual archives. User evaluation of this tool produced enthusiastic responses, and also interesting reflections on the expectations of users - ‘what do we regard as similar?’.

Keywords

• deep learning
• archive
• visual features
• audio features
• image similarity
• search