Re: [WM]: Best LSB-Steganalysis algorithm

["Jessica Fridrich" <fridrich@binghamton.edu>]

Mahdi,

RS is what started the development of reliable and accurate estimators of=   the number of LSB flippings in images. That was back
in 2001. Dumitrescu,=   Xu, and Memon explained the simplest version of RS (for masks of two  pixels -- pixel pairs) in their
ICIP 2002 paper and then expanded upon it=  in  a paper presented at IHW 2002. The most notable improvements and expansio= n  to
groups of multiple pixels come from Andrew Ker. I especially recommend=   his IHW 2005 paper from Barcelona. I would consider this
the current stat= e  of the art.

Warning! All methods above are designed for non-adaptive LSB steganograph= y,  i.e., the embedding proceeds along a pseudo-random
path through the image=   that is independent of the image content. This is a very important  assumption. LSB embedding at
adaptively selected pixels will be in genera= l  underestimated by RS, Sample Pairs, or their extended versions simply  because
the methods were designed for a different set of assumptions. Thu= s,  if your method is adaptive, the fact that RS cannot detect
it means very  little and constitutes a poor evidence of the method's security. It is li= ke  proving that the Earth is
inhabitable by looking around at the North pole=   and concluding that there is no life on earth. If it is still LSB flipping,  it is very likely that the method is highly detectable by some modified  versions of RS or some other approaches. When
designing a new stego metho= d,  you should spend at least 5 times longer testing its security that design= ing  it. I strongly
recommend testing your scheme with available blind  steganalysis methods to really see how secure or insecure the method is. You  might find this paper and the references therein useful:

New Blind Steganalysis and its Implications, M. Goljan, J. Fridrich, and = T.  Holotyak, Proc. SPIE Electronic Imaging, Photonics
West, January 2006.

The paper is available from my web site.

Finally, to see how adaptivity of embedding may fool RS and other methods= ,  but can be broken with other approaches, see

  author = {A. Westfeld and R. Böhme},
  title = {Exploiting Preserved Statistics for Steganalysis},
  booktitle = {Information Hiding, 6th International Workshop, {IH} 200= 4,  Toronto,  Canada, May 23--25, 2004},
  editor = {J. Fridrich},
  volume = {3200},
  series = {LNCS},
  pages = {82--96},
  publisher = {Springer-Verlag, Berlin},

Jessica Fridrich, Associate Professor
SUNY Binghamton
Department of Electrical and Computer Engineering Binghamton, NY 13902-6000
Ph: 607-777-6177
Fx: 607-777-4464
E-mail: fridrich@binghamton.edu
Web: www.ws.binghamton.edu/fridrich
----- Original Message ----- 
From: "Mahdi HashemiNezhad" <mahdi.hd@gmail.com>
To: <watermarking@watermarkingworld.org>
Sent: Thursday, March 02, 2006 3:51 PM
Subject: Re: [WM]: Best LSB-Steganalysis algorithm


hi all,
sorry for delay, but I'm a little busy right now; so I'll send u more inf= o  on our steganography method soon; and I'll also send
some sample photo with embedded data.

about the capacity, I tested it on 108 photos from kodak image database, = an=  the average of embedded message in which RS
could not detect (ie. detected it lower than 3%) was 22%. and for some samples the  capacity was very high.
for example in best experience, I could embed 70% data that RS could not  detect it.

yours

--Mahdi

On 3/2/06, mojtaba mahdavi <mahdavi110@gmail.com> wrote:
>
> >
> >
> > my second question: is there any steganlysis algorithm better than
> > RS?
> >
> >
> Rs is not a general steganalysis method and we know that general
> steganalysis methods can't perform detection very well.
>
> Every steganography method needs its special attack. Rs or any other
> known staganalysis method may not work on your steganography method.
> Also you should be aware of importance of amount of data that you are
> hiding. If you hide a few bits in a big carrier, nobody can detect it.
>
> Hope this may help. 


______________________________________________________________________________

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