Cryptography Reference
In-Depth Information
N
i=1
y
(f )
1
P
average luminance of pixels (ζ
i
,η
i
),
ζ
i
,η
i
, and the average luminance
N
i=1
y
(f )
1
P
of pixels (ξ
i
,ψ
i
),
ξ
i
,d
i
. Here each frame consists of 2P pixels and
y
(f )
ζ
i
,η
i
and y
(f )
ξ
i
,ψ
i
represent the luminance values of pixels (ζ
i
,η
i
) and (ξ
i
,ψ
i
)
of the f -th frame. The locations of pixels (ζ
i
,η
i
) and (ξ
i
,ψ
i
) are randomly
determined in each frame without overlapping. Small luminance changes in
these pixels result in a statistically meaningful difference if (ζ
i
,η
i
) and (ξ
i
,ψ
i
)
are in neighboring areas for each i [11]. Type (b) methods, on the other hand,
maintain picture quality directly and are usually more effective. Most of them
use the masking effects found in the human visual system [1, 7, 8]. Be-
cause picture content masks the WMs, these methods strongly embed WMs
in messy areas and weakly in plain areas. The methods select as messy areas
regions with large luminance differences and many high-frequency elements.
The quality of motion pictures is still degraded, however, because when these
methods select messy areas they consider only the properties of each frame or
still picture. They neglect inter-frame properties of motion.
7.3.2 Criteria for Watermark Imperceptibility Using Motion
Picture Properties
Our motion-adaptive embedding technique adjusts the strength of the em-
bedded WMs based on their estimated imperceptibility. It embeds them more
strongly in areas where they are estimated to be relatively imperceptible. This
section describes the criteria used for measuring WM imperceptibility that is
used to take motion into account.
Analysis of WM Imperceptibility
Analyzing the relationship between WM Imperceptibility (WMIP) and motion
picture properties (Fig. 7.2), we observed the following:
• WMIP depends on the velocity of object movement between frames the
higher the velocity, the less perceptible the WMs.
• WMIP also depends on the degree of object deformation between frames
the greater the deformation, the more imperceptible the WMs.
These observations indicate that WMIP depends not only on the prop-
erties of the still pictures making up the sequence of images constituting a
motion picture, but also on the motions of the objects in those images. That
is their velocities and their degrees of deformation. These can be estimated as a
motion vector and a deformation quantity by using a motion-estimation tech-
nique, such as block-matching. The following subsections describe the block-
matching techniques widely used for motion estimation in MPEG encoders
and present our criteria for measuring WMIP that uses these techniques.
