Digital Video Watermarking |
AlpVision does not have direct products for sale to protect video streams against fraudulent copies, but it licenses its technology to third parties. AlpVision filed US and EU registered patents, in cooperation with the Ecole Polytechnique Fédérale de Lausanne (EPFL), listed at the end of this page.
Digital video watermarking can be achieved
by either applying still image technologies
to each frame of the movie or using dedicated
methods that exploit inherent features
of the video sequence.
|
 |
The
table on the right provides a partial
list of still and/or moving picture
watermarking concepts.
One
of the primary requirements of video
watermarking schemes is their robustness
to unintentional attacks while avoiding
the introduction of disturbing artifacts.
The most important unintentional attack
on video is "lossy" compression. In
the production chain, compression
is usually applied before broadcasting
or before transferring the video to
other devices.
|
| |
Still
image watermarking |
Video
watermarking |
| X |
X |
| - |
X |
| X |
X |
| X |
X |
| X |
X |
| X |
X |
| - |
X |
|
|
|
|
The
following diagram shows the evaluation procedure
we employ to test the robustness of video
watermarking solutions against compression. Each frame of the video is first watermarked
using a 64 bit signature. The resulting
signed frames are then compressed using
an MPEG-2 encoder at different bitrates. |
 |
|
To
recover the embedded signature, the
video bit stream is first decompressed
and then each individual frame is
passed through the watermark detection
system. If the detection is perfect
all the bits of the retrieved signature
are correct and the same signature
is found in each frame. This also
means that only one single frame is
required to recover the embedded information.
However, practice has shown that in
order to keep the watermark invisible,
only very slight modifications of
the frames are acceptable.
As a consequence,
and depending on the compression bit rate,
some of the signature bits may therefore
be wrongly detected on individual
frames. An additional step based on
statistical analysis over several
frames is therefore required in order
to correct possible errors (see diagram
below).. |
|
 |
| Example |
|
|
The combination of the basic watermarking
technology with adequate statistical processing
over several frames provides an excellent solution
for invisible video watermarking that is robust
to MPEG-2 compression. The following results
show an example of the performance achieved
by this system (diagram on the right).
The first 100 frames of the "Mobile
Calendar" standard CCIR-601 test
sequence were watermarked and compressed
using 4:2:0 YUV sub sampling and three
different bit rates: 6 Mbit/s, 4 Mbit/s
and 3 Mbit/s. The graph below shows the
number of errors versus the number of
frames processed by the statistical detection
algorithm.
The graph shows that in order to recover
the correct signature at 6, 4 and 3 Mbit/s,
the required number of frames is 11, 21 and 60, respectively. |
 |
|
In other words, the signature is retrievable from less than 500 ms of video compressed at 6 Mbit/s and about 2.5 seconds of video compressed at 3 Mbit/s. It should be noted that 3 Mbit/s is a fairly low bit rate for a CCIR-601 sequence and that it creates compression artefacts that are substantially more visible than the watermark itself. |
|
The picture below is the
first frame of the signed sequence after
Mpeg-2 compression: it clearly shows the
artifacts caused by video compression (typical
effect of DCT coefficient quantization).
As can be seen, the watermarking artifacts
are very hard to detect.
The images above show the original (left) and watermarked
(right) frames extracted from compressed
bit stream.
To further illustrate this,
some examples of the original and signed
video sequences have been converted to the
standard AVI format using the following
scheme:
The Mpeg-2 bit streams "original.m2v" and "signed.m2v"as well as
the AVI movies can be downloaded by clicking
on the links below (you need a MPEG2 decoder to read the m2v files, which can be freely downloaded on the Internet): |
original.avi
original.m2v
|
signed.avi
signer.m2v |
|
|
Bit stream watermarking:
Motion vectors can also be modified without
decoding the MPEG (or similar) compressed
bit stream. This approach provides for a marking/detection
technique that has very low computational
power requirements. Details can be found
in following document US patent US6785332.
Patents:
US (US6785332) and European (EP0997042)
patents are already granted with no opposition
filed (opposition time limit has now expired)
and all fees have been duly paid. The priority date
was July 1997. The patent is entitled "Method
for marking a compressed digital video signal".
Starting January 2007, these patents are
now available for licensing or acquisition.
Please feel free to contact us for more
information.
More scientific references are available
in our Publications list. |
|
