[WM]: Re: Video Watermarking using wavelets

["Deepa Kundur" <deepa@ee.tamu.edu>]

Hi Amit,

The following papers may be of interest:

http://ee.tamu.edu/~deepa/pdf/kundur1.pdf
http://ee.tamu.edu/~deepa/pdf/kundur2.pdf

They are elaborated versions of our scheme you mention below.  The first =
paper is accepted with minor revisions pending to the
IEEE Trans on Multimedia and the second is still under review for the =
IEEE Trans on Multimedia.  They are Parts I and II of a two
part paper on video collusion resistance.  The first paper will answer =
your question:

> Suppose all frames of a video are watermarked with the same watermark
using DWT. Then, is > this video resistent to collusion
> attacks? Or is it that collusion attacks affect only videos=20
> watermarked
using spatial domain
> techniques (e.g using CDMA)? Also ,could anyone suggest a paper which
deals with
> collusion attacks, particularly against video's watermarked using
wavelets. I know
> about one paper namely "A Novel Approach to Collusion-resistant Video
Watermarking" by > Kundur, Su, and Hatzinakos.

The answer to this question depends on what you define as a "collusion =
attack").  You see, a linear collusion attack (i.e., in
which you take a linear combination of video frames to "create" a new =
frame which either amplifies or attenuates the watermark --
see first paper for exact
definition) can occur theoretically in any domain.  In fact, if the =
transform you use is linear, then it makes no difference where
you apply the attack.  A linear combination of coefficients in one =
linear transform domain is equivalent to the linear combination
of coefficients in another linear transform domain.

In this way, the linear transform domain you embed the watermark in =
doesn't matter in principle for the success/failure of such a
collusion attack. What matters are the differences in the watermark from =
frame to frame and how they relate to the evolution of
content in the host video sequence.  See the first paper above for =
details.

To answer your question, if we consider the case in which all the video =
frames have the same watermark, it doesn't matter that the
watermark is embedded in the DWT.  You can still, in part, amplify out =
the watermark (assuming the video changes from frame to
frame) by taking a linear combination of the different watermarked =
frames.  This linear combination can be taken in the spatial
domain, the DWT domain or any other linear transform domain.  However, =
since the watermark payload was embedded in the DWT domain,
it may be more practical to estimate the payload in that particular =
domain.  However, from the perspective of amplifying the
watermark (this is the signal that was added to the host to embed the
payload) which is a measure of the success of a linear collusion attack, =
it makes no theoretical difference where you do it.

Deepa
----
Prof. Deepa Kundur
Dept. of Electrical Engineering
Texas A&M University, College Station
http://ee.tamu.edu/~deepa





______________________________________________________________________________

Watermarking Mailing List - http://www.watermarkingworld.org/ml.html
To unsubscribe send email to "majordomo@watermarkingworld.org" with
"unsubscribe watermarking YOURMAIL" in the body.
______________________________________________________________________________