We sometimes meet potential customers who require a content identification system. These customers may decide to compare the two main technologies available today for that effect, Watermarking and Fingerprinting.
Watermarking Technology
The most common Watermarking technology was the one proposed by Philips via Teletrax. It is now called NexGuard, and offered by Civolution.
Forensic watermarking is a process by which a unique, invisible serial number can be added to video or audio content. The watermark is designed to remain with the content, regardless of how it might be transcoded, resized, downscaled or otherwise altered for distribution.
This system is stable, proven, and gives excellent detection. The technological principle is based on insertion of a “marker pattern” inside the video or the audio streams. The pattern is adaptive (based on the underlying video) so as not to create noise.
Digital Watermarking and Steganography
Watermarking always involves some aspect of Steganography, which is a term coming from the greek words Steganos (Covered, hidden), and graphein (writing). It is the practice of including a covert file, message, image, or video within another file, message, image, or video. The limitation of this technology is that it invariable modifies the original data that was in the Audio Video and may affect the quality of the Media itself.
Illegal Copies
Its main uses is to avoid illegal copies, in what is called “Anti-Piracy”. It is implemented either as a “Marker” to pre-release movies, to copies of Digital Cinema, by adding a Mark to STB (to track an illegal copy back to the specific STB account), or to OTT streaming sessions.
Video or Audio Content Detection
Watermarking can also be used for Content Detection, including Ad detection through either Video Content Detection or Audio Content Detection, broadcast compliance although it is somewhat an “overkill” as will be explained below, and probably a Fingerprinting Technology is more indicated..
Fingerprinting Technology
This technology consists of analysing any Video or Audio content and creating a miniature fingerprint that represents the content. In the same way that every human person has a fingerprint that is unique to that person, it is possible to reduce a few seconds of Video and Audio stream into a string that represents all the important characteristics of the original stream.
Pattern Recognition
Fingerprinting can be associated with some aspects of Machine Learning and automatic extraction of Patterns from the Video and Audio. In that sense, Fingerprinting is much more sophisticated than Watermarking as it is capable of Content Recognition.
Watermarking vs. Fingerprinting – Which technology is better?
Depending on the use case. Watermarking has advantages in precision, and capability to trace to a unique copy. Fingerprinting has advantages in being used Ex-facto, and being more cost effective. We believe that Watermarking is better for protecting and tracing specific content, while fingerprinting is better for recognizing content as a whole.
Principle
Watermarking is based on inserting a Marker into the original content before it is released. This is similar to a Police Forensic expert requiring that some Chip is inserted into all Crime suspects before a crime is committed. This chip will trigger some alert during the crime.
Fingerprinting is based on extracting a Fingerprint from any content and comparing it to the Fingerprint of another content. This is similar to a Police Forensic expert first extracting Fingerprints from a Crime Scene, and only then extracting Fingerprints from a series of suspects.
Tracing
Watermarking can insert a unique Mark into every copy of a movie, therefore an illegal copy can be traced to the original source, for example to a specific theater, or to a specific STB.
Fingerprinting cannot do that, it can only tell you that a certain content (for example a football game, or an Ad) is “very similar” to some original content (a football game that has been Marked, or a sample of the same Ad).
Watermarking has a clear advantage for Tracing.
Ex Facto
Fingerprinting does not require to insert any Marker into the content in advance. This means that if a new Ad commercial is detected in one of the TV or Radio channels, and it was never marked before, you can still Mark it now (extract a Fingerprint) and you can still search back in time (provided that you have the recorded TV and Radio shows in the Log Server archive) and find all past occurrences.
With Watermark this workflow is impossible. If the content was not pre-Marked, it will never be detected.
This is relevant in the case of an illegal Ad, or Jingle, or Promo, that you wish to detect after it has been released, or in the case of an Ad copy that was not Watermarked by mistake.
Fingerprinting has a clear advantage for back-scanning.
Precision
Watermarking is more precise than Fingerprinting. Watermarking can be inserted at practically any frequency (even 2 or 3 times per second) so it can detect even very short portions of 1 second.
Fingerprinting probability of detection increases with the length of the segment. 100% detection is obtained from about 4 or 5 seconds length.
Watermarking has an advantage here for all very short content, of less than 4 seconds.
Image and Sound quality
Both technologies will lose detection when the Video is being transcoded and modified, or when the transmission quality is very reduced.
In the case of Watermarking, one very large user of Watermarking systems wrote to us that:
“- % de Deteccion de la duracion de la Marca de Agua: En las estaciones con mejor calidad se puede llegar hasta un 95%, pero en las estaciones con peor calidad de señal se puede llegar a detectar solo (only) el 70-75% de la duración real del material.”
This translates to more or less to:
“In perfect quality transmission, up to 95% of video can be recognized using the inserted Watermarking. However if the transmission is less good then we typically detect only 70-75% of the inserted Watermarkings.”
In the case of Fingerprinting, a reduced signal can be compensated either with a longer duration of the segment or with with a lower detection threshold so that 100% detection is still maintained, but at the price of higher rate of ‘False Positives”.
Fingerprinting seems to present a better trade-off in terms of rate of detection under conditions of poor transmission.
Longer Segment
As we explained, if the copy of some Ad is exactly in the same format (resolution, and sound quality) then 4 to 5 seconds will be sufficient for 100% detection, while if the Video and Audio were tampered (different resolution, sound bitrate and format, etc.) then 10 seconds may be required for 100% detection. But the Fingerprinting will still work and detect because the Video and Audio remain with the same basic characteristics as the original (they look the same, they sound the same).
Higher False Positives
This means that in case of bad quality content, there will still be all detections, but also many cases of doubts that will be detected. The operator will have to manually inspect every detection that the system presents as “Low Probability”, but at the end of the process we will still have 100% detection.
In Watermarking, if the watermark was damaged, then the detection is lost and cannot be recovered.
Both technologies suffer from poor quality Audio and Video, but Fingerprinting is more flexible and will provide 100% detection, at the cost of higher manual intervention.
Workflow
Watermarking requires a very strict workflow. It must be inserted before the original copies are released.
Fingerprinting is more flexible. The original Ads can be released for distribution, and only later a sample file can be sent to the FIngerprinting extraction software, or even an actual recording can be extracted from the TV program itself, aven days after it has been released.
Fingerprinting has a clear advantage for Workflow flexibility.
Anti-Piracy
Actus Digital has a small number of customers using its Fingerprinting technology for Copyright protection and Anti-Piracy applications. IN most cases, given that the original content does not come with embedded Watermarking, our customers find it easier to “learn” one segment of the content (for example a series of Goals in a soccer game) and to create fingerprint markers of these segments.
Cost of Operation
Watermarking will require more human intervention during the Marking process, with more logistics, the need to obtain an original copy, to create a Watermark, and to handle the files distribution.
Fingerprinting and adding a Marker is a simple operation that takes less than one minute.
Fingerprinting on the other hand will require more human inspection to check the results of the detections. One human operator, working one shift, will be required to supervise a system detection say 20 channels, with 1000 detections per day.
Watermarking has a slight advantage on reduced operating costs.
Investment
Watermarking requires powerful servers both during the Watermark insertion process, because this process involves a modification of the original Video and Audio tracks. THis must be done on every copy of the Content that you wish to mark with a unique identifier.
Fingerprinting requires a simple software process to extract one fingerprint. It needs to be done just on one copy of the Content, any copy.
Both technologies require analysing servers that will scan each channel under inspection. But Fingerprinting can handle between 8 and 12 channels per server, while Watermarking requires more servers.
Additionally, integrating the Actus Multiviewer with fingerprinting technology enhances the monitoring process by providing a comprehensive view of multiple channels simultaneously, ensuring no content is missed.
Fingerprinting represents a smaller investment. Typically half.