Your browser is not supported. Please update it.

11 September 2014

Metadata lurks behind innovations in monetization and UI

Metadata has been around as long as broadcasting as an enabler for efficient production, archiving and distribution, but has now become a focus for innovation itself in the multiscreen era. This would not be immediately apparent to an alien visiting a major broadcasting exhibition such as IBC 2014 for the first time, given its almost total absence from any headlines. It is not mentioned once in the IBC conference program and yet crops up in some guise within almost half the presentations, having risen up the agenda in recent years.

This reflects the rising importance of metadata for advanced content search or discovery, targeted or contextual advertising and interactivity. For this reason rich metadata has become a buzzword within the field, relating to more advanced tags containing detailed contextual information at fine granularity, down to individual items, locations or scenes displayed within the video.

The power of search and potential of advanced advertising to generate additional revenues are utterly dependent on the depth, granularity and accuracy of metadata associated with that content. Apart from the ongoing challenges of implementation and interoperability the biggest hurdle by far lies in generating the metadata given that the process goes up by orders of magnitude in complexity if it starts drilling down into individual scenes. For this reason automated metadata generation has become a hot topic and will be evident at IBC from vendors such as Front Porch Digital, which will launch its Distill system for Europe at the event.

This recognizes that existing metadata will usually be woefully inadequate for advanced applications and also that the field lacks high level standards for categorizing and annotating content. Distill attempts to steer through the chaos in four stages, starting with capture and storage of the selected video files. The second stage is then to extract as much information as possible from the files from existing information such as textual data and make it available within the workflow. Thirdly the resulting unstructured data is organized and managed so that it can be accessed readily in the form of Archive eXchange Format (AXF) objects, this being an open format for interoperability among disparate content storage systems. Finally the retrieval process is enabled with the ability to search and restore from stored AXF objects.

While products such as Distill do as a good job at extracting metadata as possible in today’s video environments, there will be interest at IBC in assessing progress towards more advanced tagging through direct speech-to-text and videoshot detection algorithms, which have been coming along in laboratories over the last two years since ISO (International Standardization Organization) ratified a standard of European origin called MPEG-7 AudioVisual Description Profile (AVDP).

This provides a framework for extracting audio objects such as individual songs or even phrases and also video objects like faces from content for use as searchable metadata tags. AVDP was conceived in a project involving the EBU (European Union) in 2010, with active support and promotion from broadcasters such as RAI of Italy and VRT of Belgium, along with Austrian research institution Joanneum Research. The ISO standardization effort was then supported by Japanese broadcaster NHK and AT&T from the US, so it has some weight behind it. AVDP can be seen as part of the overall movement towards a file based IT framework that will streamline workflows with greater levels of automation and fits in with the Framework for Interoperable Media Services (FIMS), also a European led initiative.

While AVDP has gained little traction so far even inside Europe, never mind beyond it, there is growing interest in being able to generate metadata from the content itself to enable deeper and more sophisticated search and discovery as well as for triggering adverts on the fly from cues in the video. It could be said that the need for metadata itself is diminished if content can be searched and retrieved on the basis of individual audio or visual objects, but this ignores the excessive processing required, which means it could not operate in practice in real time. Therefore metadata tags have to be extracted first.