Moving image demographics: Data mining to secure value

No comments

ABSTRACT

Existing moving image collections face an onslaught of challenges to remain relevant and licensable: readiness for new colour standards and display sizes, rights-readiness and versioning in a cloud-connected world, and the ability to be found through discovery and search.

Deep Learning technology creates a wealth of new potential to drive out data to address these key issues concerning professional moving image collections. While powerful, learning systems rely on a vast and disparate set of exemplar media and ground truth data to produce strong results.

Broadcasters, producers, news organisations and archives are ideally placed to provide media for secure research, and to benefit from research results. There are opportunities to generate orthogonal resulting data, such as crosscollection media demographics and shared linked data pools.

This paper discusses the issues in the context of recent TMI/UCL Deep Learning projects and a demographics study, ‘Archive Watch’, conducted by The Media Institute and FOCAL International.

INTRODUCTION

As an industry, we already welcome a handful of resources for information-sharing, for example: the IMDB website and ID registries EIDR and ISAN provide bare-bones public data on released titles; news organisations operate business-to-business (B2B) services for data and media interchange.

JSON-LD and numerous standards provide an opportunity for cross-industry discovery and data sharing. Beyond media, industries such as air travel, fast-moving consumer goods and others have a history of cooperation to exchange B2B data within their industries, for mutual benefit.

This paper suggests there may be novel shared benefit from sharing content and data within media industries, for research purposes. The resulting ‘data mining’ enables new scientific advances and speed to market in metadata generation, and the potential for industry-wide insights and business intelligence.

Overall, this secures and enhances the value of media assets for the future. This paper builds on research projects in the fields of automated video analysis and deep learning, in particular past project Video Clarity and current project DELVE-VIDEO, and a study concerning media demographics and the opportunity for cross-collection analysis and insights, ArchiveWatch. Each of these projects was conducted with the generous support of InnovateUK, the UK’s Innovation Agency.