SysCoP for Video Streams

The transition from watermarking for still images to video sequences required several changes to the algorithm described in the previous section . The most prominent feature of video sequences is the increased sensitivity to changes introduced by the watermarking process. Even for parameter settings that minimize the watermark s strength, watermarking artifacts are visible in high-quality digital betacam video sequences.

Figure 4 illustrates the visibility problem. The block position is no longer restricted to a single image (x and y axes) but extends to the time axis, t. The modification of blocks that are close to each other in x and y as well as in t can result in flickering effects. To avoid such degradation, video streams must be processed more carefully than still images.

Homogenous areas within frames are particularly sensitive to this type of degradation as are regions containing sharp edges. Two criteria for checking blocks before actually embedding the watermark information have been introduced: edge detection and plain area detection mechanisms.

Figure Figure 5 shows DCT artifacts in edge and homogenous areas. The effects are exaggerated to make the problem visible using a still image. Figure 5 (bottom left) shows the original clipped image; Figure 5(bottom center) shows the clip marked without the check algorithms. Here, artifacts in both homogenous and edge regions are clearly visible. Figure 5(bottom right) shows the same part marked and checked for edge and homogenous blocks.

Edge detection

Numerous edge-detection algorithms are available, ranging from simple ones like the Sobel operator to more sophisticated ones, such as wavelet-based schemes. Because our algorithm aims at real-time capabilities, tools we use for edge detection must work fast. Even simple schemes like the Sobel operator have a computational complexity too high to be applied in real time. To minimize the overhead for edge detection, we use features computed within the basic algorithm cycle. The lowest frequency DCT coefficients of a transformed block can be used to decide whether the block contains an edge. The threshold value depends on the specific implementation of the DCT transform used.

Plain-area detection

The detection of smooth blocks is equivalent to a block s texture analysis. To avoid computational overhead, we use a criterion based on the block s quantized DCT coefficients. Instead of counting the number of non-zero coefficients in a transformed block's predefined region, as proposed by Benham et al., we look at the set of quantized coefficients selected for modifications. If one of the set s quantized coefficients equals zero, the block is classi- fied as a "plain block".

Robustness against MPEG-2 encoding

Despite the transparency demands in watermarked video streams, the watermark must be robust against digital TV s MPEG-2 encoding. The watermark infor- mation must survive MPEG-2 encoding, which is applied immediately before transmission. Two features of the MPEG-2 standard are very important (and challenging) for watermarking algorithms:

• The MPEG-2 encoder is very effective at removing spatial and temporal redundancy from the video stream.
• The MPEG-2 encoder keeps the video stream data rate constant. The data are compressed as much as needed to reach this goal. In rare cases, the encoder will skip complete frames.

The algorithm developed by the Security Technology for Graphics and Communication Systems department at the Fraunhofer Institute for Computer Graphics for watermarking and monitoring video streams in a TV-broadcasting environment survives MPEG-2 compression of high-quality, real-world video sequences without degrading their quality. The algorithm was designed with real-time applications in mind and has been successfully implemented in hardware, yielding a real-time marking and monitoring mechanism for high quality video signals.

Applications

The applications of digital watermarking for still images and video streams cover a broad range. The following is a list of several examples:

• Protection of copyright on semi-public image databases (e.g. at photo stock agencies). While one wishes to make the material available to as many people as possible for evaluation, the use of the material should be restricted to cases where proper reimbursement has been received. Public watermarks can also help to identify the rightful owner of an image in case a publisher wishes to use a photograph of unknown origin.
• Identification of photographs or video data (e.g. automatic retrieval of patient records based on a watermark embedded in an X-ray, marking of surveillance camera tapes with location and time codes to protect against tampering). Since the information is embedded in the data, it must be considered authentic and cannot be inadvertently or intentionally changed without degrading the data.
• Protection of intellectual property. Images and videos may contain vital R&D information or otherwise sensitive material which must be protected. If the material is marked (in the best case by a system that embeds a user fingerprint in the material before it is released to a user), there is a strong disincentive against unauthorized use or removal of data from corporate premises. This applies both to digital and analog representations.