Digital Watermarking for MPEG Video
Investigators:
Guan Yong Liang, Chua Hock Chuan, Habib Mir Hosseini,
Jin Jing, Viktor Wahadaniah, Sugiri Pranata and Huang Fengming 

Network Technology and Applications

The advent of high-quality digital compression techniques (i.e. MPEG2 and MP3) and the widespread availability of Internet connectivity have not only fuelled the proliferation of e-multimedia and e-entertainment applications, but also hastened the need for copyright protection of digital multimedia contents. The SDMI (Secure Digital Music Initiative) public challenge organized by the RIAA (Recording Industry Association of America) in year 2000 clearly revealed the music industry’s urgent desire to protect digital audio contents from unauthorized use. Major movie makers have also been actively searching for suitable solutions to prevent perfect duplication of DVD video by digital pirates. 

It is fairly well accepted now that encryption alone cannot solve all the piracy woes, and digital watermarking is the partner in fighting this battle. With digital watermarking techniques, secret information can be hidden intimately into the multimedia data so as to prevent unauthorized copying by non-compliant hardware, trace pirate distributors by intelligent software agents, or perform digital rights management functions over the Internet. In NTU, we have conducted substantial research and development works on MPEG video watermarking since 1999. These R&D efforts approach the issue of MPEG watermarking from several fronts: 

• Fast watermarking algorithms for compressed and streaming video
• Visual quality control based on HVS-adaptive embedding (HVS: human visual system)
• Watermark attacks and counter-measures
• Semi-analytical modeling and information theoretic investigation of blind watermark detection performance

Compressed and streaming video presents unique challenges such as bit rate/bandwidth conservation and real-time processing requirements to watermarking system designs. We have developed a technique for watermarking MPEG-compressed video files more efficiently than other known techniques, without sacrificing the MPEG quality and performance. Demonstration of the developed system is available in NTRC. Watermarking for streaming video is also under active development. We welcome inquiry to incorporate our technology into useful applications. 

One of the most important requirements on multimedia watermarking schemes is to ensure the invisibility of the embedded watermark. This can be achieved by adapting the embedding depth based on HVS (human vision system) considerations. Figure 1 below illustrates the clear difference in resultant visual qualities obtained using non-adaptive versus HVS-adaptive watermark embedding.

Non-adaptive Embedding                           HVS-adaptive Embedding

Figure 1: Video frames obtained after non-adaptive and HVS-adaptive watermark embedding

Watermark attack is to watermarking, just as hacking is to encryption: both are intentional and potentially professional attempts to impede the proper functioning of security mechanisms. It is well accepted now that any watermark design without attack considerations is doomed to fail some time in the future. In our team, we hence pay at least equal attention to watermark attacks/counter-attacks as watermark designs. To facilitate pirate tracing in video distribution applications, different watermarks carrying distinguishing client information may be embedded at source. If a few clients requesting for the same source data get their differently marked versions together, they may “collude” (or cooperate) to remove or weaken each other’s watermarks, leading to what is commonly termed a “collusion attack”. Collusion attacks are powerful attacks because they are capable of achieving their objective without causing much degradation in visual quality of the attacked data (sometimes, visual quality may even improve after attack.). This is illustrated in Figure 2 below, where an attacked video frame looks almost as good as the original unmarked source. 

Original (no watermark)                                Attacked (by collusion)

Figure 2: Unmarked video frame and its marked version after collusion attack

Watermarking is a relatively young technology that has been bestowed a commercially sensitive mission. Besides practical and empirical investigations, we also look deeper into its fundamental performance models and limits, in an effort to find theoretical assurance in its potentials. With such diverse ongoing R&D activities, we hope to better the understanding of the technology and contribute towards its successful adoption by the industry.