Watermarking, Fingerprinting, who cares?

The Advertising industry primarily relies on a variety of passive fingerprinting (ACR) technologies; sometimes audio fingerprints, sometimes video, and sometimes both. Fingerprinting works by listening for content. So at the time the content is presented to the consumer, it is compared by the device’s internal software containing the original metadata and a database of known recorded works. 

Fingerprinting 

• If the fingerprint  finds a match, the ACR software returns the corresponding metadata regarding the media to the client application for collection by the device manufacturer, data aggregator, or other systems platform.
• If the fingerprint of the media clip does not find a match, the media clip is directed for processing, to be added to the database. 
• The metadata associated with a fingerprint is generally managed manually by people, and supplemented by machine, or AI processing as records need to be added to the database(s).

The problem with today’s ‘alternate currency’ ad measurement is one of interoperability, meaning the commercial and programming occurrence analytics are stored across a variety of fingerprint databases. This leaves the consumer with mere fractions of the total occurrence story. Additionally, these databases are (and always have been) curated by humans after the spot has aired, and the metadata in these databases use arbitrary nomenclature and taxonomies. The various fingerprint providers each cover certain consumer touchpoints  (individual Smart TV OEMs, Proprietary detection Networks and / or consumer panels, STB manufacturers, MVPDs, and vMVPDs) but none cover the whole ecosystem.

Distinguishing amongst different fingerprinting techniques (lengths, calls to action, spoken language, etc.) poses, yet again, another significant challenge adding to the industry-wide lack of transparency.   When fingerprints alone are employed, granularity down to the version of a commercial is sacrificed. These databases can return a multitude of ad creative possibilities when presented with a fingerprint, especially as commercials get shorter. Imagine sending a fingerprint of a 3 second commercial that was cut from a 30 second and 15 second version. All three versions contain that same 3 second clip, resulting in, yet again, more ambiguity as there are now 3 possible commercials that fingerprint belonged to. 

Watermarking, on the other hand, doesn’t listen for content. A watermark is an encoded signal that is sometimes embedded in the audio, sometimes in the video, but never perceptible by consumers. When using a watermark, the watermark payload (the data that is embedded in the content) is actively pre-encoded on the content before it is distributed to media publishers, the watermark payload is usually a serial number that becomes the key to an external database, which would store metadata about the content. When the content is presented to the consumer, it is run through a decoder that is embedded in the device, for detection.   

• The watermark payload returns the corresponding metadata regarding the media back to the client application for collection by the device manufacturer, data aggregator, or for other data processing needs.

This process makes the  detection of a watermark in the content more efficient and accurate than fingerprinting. Since the watermark contains a predefined serial number that is embedded at a consistent location in the content, the detection capability uses less processing power on the receiving device, than a fingerprint, which requires the entire content to be processed for comparison to the database of known works. In addition, the pre encoded nature of the watermarking process shifts control of  metadata management to the original content creator. This removes the need and risk that comes with third-party management of the metadata, obsolete & outdated fingerprint databases, after-the-fact detection reports and potential human error. 

The fact that each watermark payload is a unique serial number, with distinct associated metadata, underpins the certainty that individual versions of the creative will have distinguishing metadata.

Watermarks are not easily removed from content and  are currently deployed for selected granular ad reporting and verification purposes. Think of them as digital insurance policies embedded into content/ad creative. A universal deployment of watermarks would improve interoperability, and thus the ability to aggregate across vendors and platforms. This will enable all sorts of “census based” analytic applications ranging from the underpinnings of excellent measurement, second screen interactivity functionality, attribution, and other applications for both entertainment content and Ads.

As an industry, we must adopt technologies that are complementary to one another, and start with a baseline that underpins granularity and precision. The IAB Tech Lab recently released the Ad Creative ID Framework (ACIF), which defines persistent creative identifiers being standardized and enabled across linear and digital media platforms, Fingerprinting does not easily lend itself easily to implement ACIF, while watermarking by design facilitates the granularity necessary to create a precise connection to an ACIF compatible identification system, like AD-ID.

As an industry we must begin to look at watermarking as a critical baseline, and fingerprinting  as complimentary  technology can add depth and context to the accuracy, speed, and granularity that watermarking enables.

Future blog posts will go into depth on Audio, and Video Watermarks, and the innovation that they enable.