view Fotopoulos/README_VASSILIS @ 5:8191b855cd5e

Added tag v0.5 for changeset 9cf6ec970448
author Peter Meerwald <pmeerw@cosy.sbg.ac.at>
date Tue, 14 Aug 2007 21:13:29 +0200
parents be303a3f5ea8
children
line wrap: on
line source

These codes use the DCT,Hartley and Subband DCT
Transforms for watermarking purposes.All schemes
are blind,no original image is used for detection.
Each casting module is accompanied by a testing 
module.Supposing that the watermark key that you
select during casting is in the range 1-1000,the
testing module will test all the keys in this range
to produce the well known similarity diagrams used
by Cox, Piva and many more with the peak being the
proof of the watermark existence.

I do not include a detection module because this implies
the use of certain thresholds. Although the casting
methods are almost standardized (multiplicative formula)
there are still questions about this thresholding but
the testing modules can be easily changed to fit this
purpose.

The DCT scheme does not use the visual masking improvement
that Barni and his team suggest in one of their later works.
In all casting programs we assume that the coefficients are
diagonaly scanned,and ordered as shown in the following example.

----------------------
| 1| 3| 6|10|15|...
----------------------
| 2| 5| 9|14|...
----------------------
| 4| 8|13|...
----------------------
| 7|12|...
----------------------
|11|...
----------------------

which is quite simple comparing to the zig zag scanning 
pattern that we know from the JPEG standard but does not
affect at all the idea that we have of the middle 
frequencies.Also the user should take care that

starting_coefficient+number_of_coeffs_to_change<(N*N)/2

which means that we shouldn't exceed the matrix diagonal.
To do so the scanning scheme should be changed accordingly
but this doesn't seem important because in all of the
schemes we don't get out of the middle frequencies coefs.
With correctly selected parameters,the schemes perform in
the same way as if the coeffs were zig zag scanned.

In the Subband DCT version,the first set of questions
about starting coefficient,number of coefficients to
alter and alpha parameter, refer to the LH,HL and HH
band while the second set of questions sets the parameters
about the LL band only.

All schemes should not be used with the same set of parameters
because each of the transforms, possess certain specific
properties. This should be kept in mind for testing purposes.
To use the right set of parameters please refer to corresponding
bibliography.
All code has been tested in Visual C++ v6.0

Have some nice tests...
Vassilis Fotopoulos

for more info,ideas or points of discussion
email vfotop1@physics.upatras.gr

Repositories maintained by Peter Meerwald, pmeerw@pmeerw.net.