|
The IBM DataHidingTM technology is
based on signal processing and statistical analysis techniques
for realizing highly transparent, reliable, and secure watermarking.
It allows a content owner to embed an invisible message directly
into the digital video data.
Since the embedded message is tightly coupled with the video data,
it cannot be removed without damaging the visual quality,
and is expected to be carried with the video content
whatever the distribution path, whatever the data format, and
whatever the signal processing.
In IBM DataHidingTM, the embedding process consists of modulating a quasi-random
pattern with multiple bits of an invisible message
and adding them to the luminance components
of each pixel in the frame or field of digital
video data. The modulated quasi-random pattern
is modified locally and collectively by using a
perception model, which takes account of the
visual effect of embedding by analyzing the local
image and embedding patterns.
This automatic and adaptive approach
is designed
to ensure maximum detectability of
the embedded
data, while simultaneously ensuring
the absence
of perceptible effects on the quality
of
the embedded image.
The detection process consists of modulating the input video data
with a quasi-random mask, accumulating the modulated data, and extracting
the embedded message by statistical inference.
There is no need for the embedding pattern and detection mask to be exactly the same,
since the accumulation and statistical inference are applied.
Intentional use of the embedding pattern that is different from the detection mask increases
the security of watermarking.
The major advantages of the IBM DataHidingTM
technology are summarized below.
- It is very flexible in terms
of optimizing the overall trade-off among
the competing requirements of visibility,
data capacity (payload), detection reliability,
and security.
- It allows automatic and real-time
embedding with high transparency.
- It shows high survivability with respect to:
- normal studio video processing techniques such as filtering
- MPEG encoding and decoding; the survivability against CBR, VBR and re-MPEG2 encoding has been successfully tested.
- digital-to-analog conversion
- noise addition
- horizontal and vertical shifting
- arbitrary scaling (aspect ratio is not necessarily
constant)
- cropping
- exchanges in time-series
- It is compatible with real-time
detection from both baseband and MPEG2-encoded
contents without decompression.
- It has a theoretical basis for reliable detection.
- It provides an adaptive frame accumulation detection method,
which automatically adjusts the detection time window to extract the correct
message from a weak or corrupted signal
with a predefined low false positive error ratio.
- It provides an extremely large number of possible pseudorandom masks,
to prevent any dictionary attack.
DataHiding Home Page
|
