You are here

Multimedia Classification

BBC R&D’s Multimedia Classification team are investigating ways in which different kinds of metadata can be extracted from programmes within the BBC archive and used to assist content navigation. With the BBC aiming to digitise and open up the archives to the public by 2022, the need for more effective search methods is imperative, both in terms of helping users find what they’re looking for and recommending content the user's didn’t know about.

To this end, we are looking at extracting and utilising both semantic and affective metadata – identifying not only the topic or factual content of a programme or scene but also its mood or emotional value. For this we are developing techniques in various fields such as vision and audio processing, for example to identify faces or the mood of background music along with research in speech recognition and text analysis.

Our final area of research is that of machine learning where we aim to map the extracted metadata to semantic and mood based values. This will allow viewers to find programmes not only based on more search strategies (programme title, actor etc), but also in novel ways which are more intuitive to how humans classify and respond to media.

Presentation Type: 
Presentation Audio: 
Audio Size: 
Presentation Visual: