Forensic watermarking is gaining a lot of interest as a way of preventing content leakage and tracking leaked streaming content by embedding watermarking IDs or codes. The rise of OTT platforms, particularly during the Covid-19 pandemic, has contributed to this interest. A forensic watermark, also known as a digital watermark, is a code or a set of characters embedded into the digital document, video, audio, image, or program which enables the unique identification of the content creator and its authorized user.
Forensic video watermarking is further divided into manifest-based or A/B watermarking, and bitstream-based watermarking. In the A/B watermarking method, the pixels are modified in the transcoder and the content undergoes some pre-processing to identify the modifiable pixels and modify them without compromising the quality of the video. Pre-processing can be done by CLI preprocessor, preprocessing libraries, or a SaaS packaging service, based on the preference of the content owner. This pre-processed data is then fed into the transcoder, where the pixels are modified.
The assets are segmented into chunks, so that the playouts have a unique pattern of As and Bs. Different values (A/B or 0/1) are inserted in the original video frames and the output is given as two sets (A/B) on the encoded video. In this way, two content copies are used to create a unique manifest for each subscriber session. Finally, the packager combines the video segments from each of the two copies to create a unique manifest. The watermark on this unique manifest can then be used to identify the original customer of an illegally shared piece of content.
The advantage of A/B watermarking is that it is secure and hard to break, thereby offering an effective solution for DRM video protection. The downsides are issues with scalability and complexity, since two copies need to be stored, and each subscriber must receive a unique manifest. It is also possible for hackers to combine the segments from two different streams and thereby create and share a unique false copy of the asset. This is known as a collusion attack.