How AI can save you from drowning in the archive
Media versioning or mastering is central to various Editorial, Post-Production, and Content Distribution workflows. Digital technology aids, in the form of media scanning, restoration, archiving, and storage, have accelerated these workflows ten folds. This has further fuelled the motivation to repurpose content for different distribution channels to capture the growing revenue horizons. It is particularly essential now during COVID-19 time when new content production is severely impaired.
Archive libraries are growing at a breakneck pace. Content distribution workflows can result in deep archives with many different distribution versions. Especially for International Distribution purposes, archives can have hundreds of versions of the same content. For different syndication or theatrical distributions, editors create versions, including proxies, direct intermediaries, and master packages. Each version has unique requirements such as alternate edits, aspect ratios, resolutions, color spaces, captions, commercial breaks of different lengths, channel brandings, compliance edits, texted overlays, and many other variations.
Working with these versions can have productivity challenges for an archivist or an editor, which can lead to growing inefficiencies in media versioning management workflows. For instance, doing a spot proxy check on a restored file from an archive can be a daunting task, requiring a search through hundreds of archived versions. It’s a waste of time for an editor to find the best version among the numerous versions in a deep archive. For instance, to create a new DCP master from a Digital Intermediate, an editor must validate the quality of different intermediaries by comparing edits, cuts, color space differences, text differences, dead pixels, etc. These are time-consuming tasks and may keep an editor away from actual creative work. No archivist or editor would like to be in such situations resulting in productivity inefficiencies.
Similarly, when an editor needs to re-master an older version of a master into a higher resolution, they need to restore only matching dailies footages from the archives so that these selected dailies files can be scanned in higher resolution to create a new master. For this purpose, editors need to compare every segment in reference master cut with original uncut dailies sources. In this case, both comparison and identification of select high-res dailies from archives can be daunting. To find matched dailies sources, an editor will need to compare a reference master with all the original production dailies for close similarities, including color spaces, VFX differences, green screen differences, zooms, crops, and other image structural changes. To conform a few minutes of final content requires sorting through many hours of dailies. For instance, conforming a forty-minute final cut with 20 hours of dailies will take up to three or four person-days.
But suppose there is an easy and automated approach to compare different masters. In that case, it can help editors purge duplicate versions and compare different versions to help preserve only quality versions in the archives. For re-mastering content, editors can use automated video comparison to conform final cuts with multiple source files in just a few hours. A comparison will tell which files precisely need restoration. An automated video comparison tool will expedite working with different versions. It will help preserve only desired content, instead of multiple duplicate or edited versions. It will speed up syndication distribution workflows by restoring quality versions and directly initiating edits to create new masters. As an outcome, it can boost not only operational productivity but individual performances as well.
Video comparison is not a very straightforward task. Any attempt to match pixels frame by frame is a substantial computational exercise and time and cost critical operation. It is about estimating key features at every frame and qualitative comprehension, such as identification of color space differences, different resolutions, zooms, crops, etc. This makes automated video comparison a complex and challenging effort. Using Artificial Intelligence (AI) and leveraging computer vision methods, it’s possible to compare videos on a broad spectrum of automation complexity. Editors can compare videos by leveraging perceptual hash techniques, or abstracting visual information as key features using different variants of image feature detector algorithms. More sophisticated custom methods may be required to enable the qualitative analysis of the video comparison. For all of the above possibilities to fall into place, it’s critical to enable video comparison in a most optimal and usable manner so that it’s a seamless use of advanced technologies to manage your media archives easily.
