Cryptography Reference
In-Depth Information
Statistically adaptive detection [6]: It is common in conventional watermark-
ing to embed identical watermark patterns in different regions. That is
known as redundant coding and is used to improve resistance to video
processing without degrading picture quality. The watermarks can then
be detected by accumulating those from each region of interest. Some
video processing methods remove watermark signals from specific regions
of the frames. This degrades the combined signal of the accumulated wa-
termarks and thus reduces the detection ratio. Our statistically adaptive
detection technique sets up a scale for estimating the bit-error rates of
the WMs for each frame region by using statistical properties of motion
pictures and uses those rates to identify the best regions to detect the
watermarks and thus improve the detection ratio.
In the following section, we describe the above video watermarking meth-
ods in detail and show their effectiveness.
7.3 Motion-Adaptive Embedding Using Motion
Estimation
Because video watermarks must not degrade picture quality and must sur-
vive video processing, the properties of the pictures are generally taken into
consideration so that the watermarks can be allocated adaptively. Picture
quality degradation is avoided by embedding watermarks sparsely in plain
areas, where they are easily perceived; survivability is ensured by embedding
them heavily in messy areas, where they are hard to perceive [1, 7, 8]. Conven-
tional watermarking methods have trouble satisfying these conflicting quality
and survivability requirements simultaneously. That is because they consider
only the properties of each frame. That is of each still picture and neglect the
inter-frame properties of motion. In this section, we explain how watermark
perceptibility is affected by inter-frame motion and distortion and describe
our adaptive watermarking technique which employs motion estimation.
7.3.1 Conventional Methods
Conventional methods for maintaining picture quality can be classified into
two types:
(a) those that reduce the quantity of embedded WMs by making a small
number of WMs robust enough to be reliably detected [9, 10, 11], and
(b) those that embed WMs where they will be less perceptible [1, 7, 8, 12].
Type (a) methods maintain picture quality by reducing the luminance
change while maintaining WM robustness. One such method is an improve-
ment of the patchwork algorithm [11] reported by Bender and coworkers [13].
It embeds WMs by creating a statistically meaningful difference between the
