Re: [WM]: a question about cox's algorithm

[anurag Mishra <anurag_cse2003@yahoo.com>]

Hello,

 

You seem to be making a mistake by choosing green channel which is supposed to be highly vunlerable for image processing operations such as watermarking. Instead go for blue channel. This is my suggestion only.

 

Moreover, the value of p has to be optimised. I took p=0.1 for blue channels, results are quite good, both for embedding as well as extraction.

 

I also could not understand the logic of taking a = -1 and +1 only as 'a' has to be the real value from the pseudo random number sequence generated as watermarks followed by statistical distribution such as Normal Distribution with mean 0 and variance 1. Thus, all a values will be in the range -1.0 - 0.0 - +1.0 including endpoints and not simply -1 and +1 only.

 

Anurag Mishra

yanmiaobest <yanmiaobest@163.com> wrote:

Hi,
   thank you very much for you reply!

   my method is this:
              
            divid the original image into 8*8 subimages, and apply dct transform to the green channel of each block. And chose some coefficient to inject the signature.
            but I encounter a problem.  Injecting into the low frequency area is perceptable, injecting into the high frequency area is subjected to attack, is that right? but wharever area I inject the signature, it is more or less perceptible. My formula is:   x = x * ( 1 + a * p ), where  a = -1 when signature is 0, a = 1 when signature is 1. and p = 0.3. x is original image tranformed by dct. I could use smaller p value to avoid this, but that makes the extraction very difficult. Smaller p value means chaning the x very little. My extaction method is: sig = 1 when Xw > Xb, otherwise sig = 0. Xw is the watermarked image transformed by dct, and Xb is the original image transormed by dct( all applied on the green channel). when p is too small, this method sucks, because after injected the watermark, I have to transformed it into spatial domain using inverse dct, but many value is the same as the original image after inverse dct, so I can't extract the watermark.
           please help me
best
 regards

 

 

 

What do you mean by variance of cox's algorithm?=20    =20    The answer to your
question is : It
 will depend upon the gray level of =  your image. For ex., if the image is represented by 8bpp  (256 color) then according to Cox,
the 63 coefficients of every block =  will be sorted in descending order, barring DC coefficient  (0,0), select first few largest
coefficients and insert the random =  number according to formula given in paper. The random numbers  will be real numbers with mean
0 and variance 1. After embedding them =  the pixels which were integers initially will be converted to  real. They will be rounded
off to its nearest values. This does not =  result in much difference in perceptible quality and SIM or PSNR  values.    =20    Good
Luck    =20    Anurag Mishra    =D1=D6=ED=B5 <yanmiao06@mails.gucas.ac.cn> wrote:    hi all  I am new in this field. And I am
about to impelment a variance of cox's =  algorithm. The image is divided in 8 * 8 block, and dct  transform is applied to each
block, then insert watermark into each =  block. My questions is
 this:  1. Dose each pixel in a 8 * 8 block can only be inserted one bit, which =  means a block can at most inject 64 bits
infomation?  2. when trasformed using dct, the values are real numbers, but after =  inject the watermark, I have to round real
numbers into  integer, because image gray-