Forensic Watermarking

1. Overview
2. How does Forensic Watermarking work?
3. Does Forensic Watermarking affect video quality?
4. What happens in the event of a leak?
5. Working with Forensic Watermarking

Overview

Forensic watermarking is a security technology intended to deter the leaking of video assets throughout the various life cycles of a production. 

It’s accomplished by embedding an invisible watermark into the videos that acts as a unique identifier to the specific user viewing it. The forensic watermark then follows the asset across any environment where the viewer watches it; from browsers and phones to tablets and set-top boxes. In the event of a leak, the unique watermark then serves to aid us in discovering the source of the pirated material.

There are two other forms of visible watermarking in addition to the invisible type that comes with forensic watermarking - overlay and burn-in. These display information such as user name, IP address, and production on the assets where the viewer is able to see it. Visible overlay watermarking simply displays that information over the video as it’s being viewed, whereas visible burn-in watermarks are encoded into the video at point of play. We provide visible burn-in watermarking on both streaming and downloaded assets. 

Forensic watermarking’s advantage when compared to visible overlay is that the watermarking is encoded into the file itself, and not as information that the file could be pulled out from under and severely hinder any ability to successfully investigate and discover the source of a leak. Compared to visible burn-in, forensic watermarking gives viewers a full, clean image to watch without the presence of any other distracting information while also being able to play a role in detection when visible watermarks have been tampered with or masked.

Any of those options are valuable deterrents to the pirating of assets. If someone sees or knows that personally identifiable information is attached to a file, they’re less likely to try stealing it, but only forensic watermarking is able to maintain both a robust level of asset security and image quality that delivers at all stages of modern film and television production.

How does Forensic Watermarking work?

We partner with ContentArmor to deliver forensic watermarking that maintains our high standards of asset security, and the process starts after an asset has been uploaded.

Once that’s complete, the asset goes through a stage called profiling. This analyzes and identifies portions of the asset’s bitstream - the ones and zeroes it’s made of - where imperceptible forensic watermarking changes can be made to the video. Each combination of changes is called an Identifier, and they’re all generated, saved, and copied into a database to be used for potential future investigations.

The identifiers are also kept for future transmission within the asset at the point of play. When a viewer logs in and starts watching a forensically watermarked asset, the changes identified during processing are then embedded into their stream as it plays back. If the asset is downloaded for offline viewing, the identifiers are encoded into the file upon download.

While this does incur an upfront time investment at the upload/processing stage, there is almost no added delay upon playback when watching forensically watermarked assets. 

Does Forensic Watermarking affect video quality?

The changes being applied to videos through this process are visually imperceptible. Picture quality analysis of ContentArmor’s high fidelity forensic watermarking on assets has been assessed by studio-affiliated golden eyes; sight experts trained to perceive the smallest imperfections or impositions in an image. 

What happens in the event of a leak?

In the event that an asset ends up in the wild, an investigation can be performed to determine its origins by utilizing the forensic watermarking and a list of suspected source assets. The watermark itself is capable of surviving a range of signal processing attacks, as well as conversion from digital-to-analog or analog-to-digital.

The first step is to notify us of a suspected leak. The best way to do that is via support@moxion.io with the word, "urgent" anywhere in the subject or body of your email. Upon notification, we'll assemble a response team that includes an engineer and a lead engineer. Then we'll need a list of source assets from the leak that you suspect have come from our apps as well as the pirated copy of the file. From there, we'll download any clips which may have been pirated, as well as their Content Armor metadata. All of this information is uploaded into a secure portal at Content Armor where they'll be used to determine the forensic ID of the leak's source.

At this point, we'll know if there's any need to share the original source asset with Content Armor. If we don't, then we can cross-reference the forensic ID with user IDs and access logs from our end to determine the source of the leak.

Working with Forensic Watermarking

While forensic watermarking has the benefits of being imperceptible, traceable, and robust, there are some factors to take into account when using it.

• Processing time is about doubled, and should be factored in when uploading any video assets. The forensic watermarking we employ is only applied to the visual portion of videos and audio remains untouched by forensic watermarking. 
• Note that original files are kept as-is upon upload, and do not get forensically watermarked. It is only proxy files that are encoded with forensic watermarking
• For anyone who has the permissions to remove watermarks (Admins or Downloader (no watermark), it should be noted that this does NOT apply to forensic watermarking. If forensic watermarking is enabled, regardless of your watermarking settings all proxies will be forensically watermarked. If you are downloading a proxy it will always have forensic watermarking applied. 

There are also some factors that can limit the effectiveness of forensic watermarking.

• An asset length of about two minutes provides the ideal minimum for the most robust watermarks. Shorter clips will still be watermarked, but that decreased runtime means less data to work with in embedding identifiers. That said, a short clip with many visual changes is as good as most longer ones.
• Assets with higher amounts of visual variety allow for profiling to produce more and better identifiers than those with predominantly static imagery. Clips from Editorial or VFX workflows (including previs) are ideal, and about half of those that come through from Immediates are well-suited to this forensic watermarking. 
• Some clips may not be able to give us automated detection when performing forensic investigation. In such an event, we may need to provide a copy of the suspected original asset to ContentArmor for further inquiry.