Digital watermarking in a perceptually uniform domain

Details Digital watermarking in a perceptually uniform domain Digital watermarking in a perceptually uniform domain

2000-08-23

2004-09-14

Au, Amelia M. (Department: 2623)

Image analysis

Applications

C382S248000, C382S274000

Reexamination Certificate

active

06792129

ABSTRACT:

FIELD OF THE INVENTION
This present invention relates to digital watermarking of digital media, and more particularly to a method and apparatus for watermark insertion and detection in a perceptually uniform transform domain.
BACKGROUND OF THE INVENTION
Digital watermarks have been proposed as a means for copyright protection of digital media such as images, video, audio, and text. Digital watermarking embeds identification information directly into a digital media object by making small modifications to the object itself. A companion watermark detector can extract this “signature” from the watermarked media object. The extracted signature can be used to identify the rightful owner and/or the intended recipients, as well as to verify the authenticity of the object. The signature can also be used to embed some other useful information such as copy control information and parental control information.
For most applications, two basic desirable criteria for a watermarking scheme are perceptual invisibility and robustness to intentional/unintentional attacks. The watermark should be perceptually invisible, i.e., it should not noticeably interfere with the perceivable quality of the object being protected. The watermark should also be robust to common signal processing and intentional attacks. Particularly, the watermark should still be detectable even after common signal processing operations have been applied to the watermarked image.
The dual requirements of perceptual invisibility and robustness, unfortunately, conflict with each other. That is, the former suggests that the amount of watermark energy inserted into the object should be minimized, while the latter suggests the opposite. One of the fundamental issues in digital watermarking is thus to find the best trade-off between imperceptibility and robustness to signal processing.
One way to balance perceptual invisibility and robustness is by incorporating explicit human perceptual models in the watermarking system. The perceptual models provide an upper bound on the amount of modification one can make to the media content without incurring a perceptual difference. A watermarking system can operate just within this upper bound to provide the maximal robustness to intentional or unintentional attacks, given a desired perceived quality.
For example, when a watermark is applied to a visual object intended for human viewing, the watermarking system can exploit various properties of the human visual system (HVS). That is, some researchers have attempted to hide the watermark where it will least be noticed by a human viewer. In U.S. Pat. No. 5,930,369, entitled “Secure Spread Spectrum Watermarking for Multimedia Data”, Cox et al. teach a method that operates in the frequency domain, distributing the watermark within the n largest low-frequency (but not DC) transform coefficients. Cox et al. teach that the amount of watermark signal inserted into a particular coefficient can also be made proportional to the value of the coefficient itself.
Other researchers have taught the use of explicit HVS models to vary watermark energy. Podilchuk and Zeng suggest such a system in “Image-adaptive watermarking using visual models,”
IEEE Journal on Selected Areas in Comm
., special issue on Copyright and Privacy Protection, vol. 16, no. 4, pp. 525-39, May 1998. Podilchuk and Zeng make use of frequency sensitivity, luminance sensitivity and the self-masking effect of the HVS to adaptively control the amount of watermark energy to be embedded into different transform coefficients/areas of the image. They suggest incorporating perceptual models in the watermarking system by deriving a just-noticeable-difference (JND) threshold for each DCT/wavelet coefficient, and using this JND threshold to control the amount of watermark signal inserted into each coefficient.
FIG. 1
illustrates, in block diagram
20
, the watermark insertion scheme proposed by Podilchuk and Zeng. A frequency-based transform (e.g., a block based discrete cosine transform (DCT) of an original image {x
i,j
}) produces a frequency-based representation of a digital media object {X
u,v
}. JND calculator
24
uses frequency sensitivity, luminance sensitivity, and contrast masking models to compute a JND value J
u,v
for each X
u,v
. Watermark embedder
26
receives {X
u,v
}, {J
u,v
}, and a watermark sequence {w
u,v
}. For each component X
u,v
, embedder
26
produces a corresponding output component X*
u,v
using the formulation:
X
u
,
v
*
=
{
X
u
,
v
+
J
u
,
v
⁢
w
u
,
v
if
⁢
 
⁢
X
u
,
v
>
J
u
,
v
X
u
,
v
otherwise
Finally, frequency-based inverse transform
28
inverts X*
u,v
to produce the watermarked image x*
i,j
.
FIG. 2
illustrates, in block diagram
30
, the watermark detection scheme proposed by Podilchuk and Zeng. The original image x
i,j
is input to a frequency-based transform
22
and JND calculator
24
identical to those used in
FIG. 1
, producing X
u,v
and J
u,v
as described above. The potentially-watermarked image x*
i,j
is input to an identical frequency-based transform
32
, producing a frequency-based representation of that image X*
u,v
. Adder
34
subtracts X
u,v
from X*
u,v
, producing a difference sequence w*
u,v
that represents a potential watermark sequence (and/or noise). Correlator
36
correlates the original watermark sequence w
u,v
with the difference sequence scaled by the JNDs, w*
u,v
/J
u,v
. Comparator
38
examines a resulting correlation figure, declaring the existence of the watermark if the correlation figure exceeds a selected threshold.
In another work, the watermark embedding takes place in the spatial domain, but the perceptual model is used in the frequency domain to shape the resulting (watermarked) coefficients to make sure the modification to each coefficient does not exceed the perceptual threshold. M. Swanson et al., “Transparent robust image watermarking,”
Proc. Inter. Conf. Image Proc
., vol. 3, pp. 211-14, September 1996.
The need for watermark detection without the assistance of an original data set exists in several circumstances. First, as a content provider, an automated search for your watermarked content, e.g., over the Internet, may be practically limited to a search without the original, because the automated searcher may have no good way of determining the corresponding original for each file examined. Second, in some circumstances it may make sense to add the watermark at the time the media object is first captured or created, in which case no “un”-watermarked copy exists. Likewise, for security or storage efficiency, it may make sense to destroy the original copy. And when ownership is to be proven, use of an “original” object may be disallowed in order to avoid questions that can arise as to whether the “original” was possibly derived from the “watermarked” object.
In each prior art method, the watermark embedding and detection are implemented in either the spatial pixel domain or a linear transform domain. As a result, the amount of watermark energy to be embedded into each spatial pixel or linear transform coefficient varies from pixel to pixel, or from transform coefficient to transform coefficient. This, in general, makes optimal watermark detection difficult to design and implement in these domains. These problems are compounded when the original image is not available to assist in watermark detection.
SUMMARY OF THE INVENTION
Whereas the prior art has focused on ways to add the appropriate level of watermarking on a per-sample basis, the present invention takes a different approach to watermarking. The basic concept underlying this approach is watermarking/detection in a transform space that allows the same level of watermarking to be applied to all samples. For instance, in one embodiment, a watermarking system first nonlinearly transforms the original signal to a perceptually uniform domain, and then embeds the watermark in this domain without varying the statistical properties of the watermark at each sample. At the watermark detector, a

Affiliated with

Also associated with

Rating

Digital watermarking in a perceptually uniform domain does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Digital watermarking in a perceptually uniform domain, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Digital watermarking in a perceptually uniform domain will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3222264