RE: [WM]: Want to get some infor about watermark

[Robert Hsin <rhsin05@yahoo.com>]

Interesting questions.

1. To recover the watermark from a scaled version of the  stego image, you have to scale it back first (just as Cox et al's
experiments). In this way, you can still extract the watermark, even if only a quarter of the stego image is used. If you don't
scale it back, definitely you are not able to detect the watermark which is based on DCT.

2. When you use a HVS model, PSNR is no longer a good metric of image quality. You might want to try weighted PSNR (WPSNR), which is
more reasonable.

3. If you use HVS-based watermarks, you should consider applying an optimal detector, rather than the often used coorelator.

Good luck

-Robert



--- Nizar Sakr <nsakr@site.uottawa.ca> wrote:

> 
> Hi everyone,
> I'm currently working on an image watermarking technique, and it's 
> based = on Cox et al.'s scheme.
> In brief the algorithm divides
> the image into 8*8 blocks and I use an HVS to embed into the 
> perceptually significant components of the image (in the DCT domain).
> My algorithm is very robust against JPEG and cropping, however it is 
> very weak against scaling = attacks, I can't detect the watermark when 
> scaled by 50% (256*256 --> 128bt128), = does it have to do with the 
> fact that I'm dividing my image into 8*8 blocks?
> = Any suggestions? (I'm using the baboon image for
> now)
> 
> Another brief question, I'm choosing a specific region to embed my = 
> watermark (within an 8*8 blocks), for some images, I'm able to 
> generate a = watermarked image with excellent PSNR, for other images 
> though, I get a very low = psnr and the watermark is no longer 
> detectable (even without the presence of = any attacks). What could be 
> the cause of that? I don't think it has to do = with the HVS, it's 
> something related to the DCT domain that I don't quiet understand. How 
> would I be able to choose the appropriate area in any = image that 
> would insure that a watermark will remain detectable (again under = no 
> attacks)?  Any papers discuss these issues?
> 
> Thank you in advance,
> Nizar
> 
>
-----------------------------------------------------------------
>  
> Nizar Sakr
> M.a.sc. Candidate
> (http://www.site.uottawa.ca/~nsakr)
> School of Information Technology and Engineering University of Ottawa  
> 800 King Edward Ave. SITE 3010
> Tel: 1 (613) 562-5800 ext.2197


______________________________________________________________________________

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