Integrating forensic watermarking into adaptive streaming workflow

No comments

ABSTRACT

Per-session forensic watermarking is a process of embedding a unique identifier for each streaming video session into the video content.

This identifier will remain in this content, even if the content undergoes multiple cycles of camcording, filtering and transcoding. Forensic watermarking in adaptive HTTP streaming service is a two-step process, and thus the standard workflow is altered in two disjoint points.

In the first step, preprocessing, some segments are duplicated and imperceptibly marked, creating two variants of the same segment. These variants are perceptually identical, but contain different bits of information.

At the second step, a unique sequence of A and B variant segments is generated per each session, and this list is translated into the manifest format of choice, such as DASH MPD or Apple HLS. This stage is referred to as embedding.

This paper will discuss capacity, storage and visual quality trade-offs in selecting a workflow stage at which pre-processing is performed. The two widespread options are a preprocessing step in baseband video, prior to encoding or encoder independent in the compressed-domain. This paper discusses and quantifies ways of reducing capacity and the storage impact of forensic watermarking in both approaches.

It will further explore robustness and security implications of the two approaches with respect to different potential vulnerabilities, such as collusion. Lastly, this paper will cover approaches for enabling per-session embedding compatible with existing DVB and DASH-IF compliant MPEG DASH clients.

INTRODUCTION

Digital rights management (DRM) and conditional access systems (CAS) are often seen as the only content security mechanism needed for premium content.

However, the main goal of a DRM system is to prevent unauthorised playback of the content. It is not intended to protect from unauthorised distribution of the presented content. Content can be recorded at playback from HDMI output or by camcording. Content can also be obtained via a DRM failure, such as a key compromise or a wholesale compromise of the DRM system.

When unencrypted content is distributed, rights owners have little or no visibility into its distribution. Forensic watermarking allows precise identification of the viewing session from which this content was obtained in its unencrypted form. This is why watermarking is increasingly viewed as an important content security mechanism for premium content, such as early release and UltraHD movies. The MovieLabs Enhanced Content Security specification explicitly requires robust forensic watermarking as a part of its content security suite of technologies.

Watermarking embeds identifiable information (watermark identifier) into video signal in a way that (a) is imperceptible to a human viewer, (b) allows precise identification of each viewing session, and (c) is robust enough to survive multiple iterations of transcoding, image processing, and camcording, while maintaining its veracity.