Journal of Software - Academy Publisher
Journal of Software - Academy Publisher
Journal of Software - Academy Publisher
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
852 JOURNAL OF SOFTWARE, VOL. 6, NO. 5, MAY 2011<br />
'<br />
LL<br />
CF<br />
i, j, s, t<br />
'<br />
i, j, s, t<br />
M M<br />
⎧ α1<br />
⎪LLi,<br />
jst , , + 2 ∑∑LLi,<br />
jst , , if wst<br />
, = 1<br />
⎪ M s= 1 t=<br />
1<br />
= ⎨ M M<br />
⎪ α1<br />
LLi, jst , , − LL 2<br />
i, jst , , if wst<br />
, = 0<br />
⎪ ∑∑<br />
⎩ M s= 1 t=<br />
1<br />
(2)<br />
M M<br />
⎧ α2<br />
⎪CFi,<br />
jst , , + 2 ∑∑ CFi, jst , , if wst<br />
, = 1<br />
⎪ M s= 1 t=<br />
1<br />
= ⎨ M M<br />
⎪ α2<br />
CFi, jst , , − CF 2<br />
i, jst , , if wst<br />
, = 0<br />
⎪⎩<br />
∑∑ M s= 1 t=<br />
1<br />
(3)<br />
The parameter α 1 and α 2 are the embedding<br />
intensities and CF maybe one <strong>of</strong> HL and LH . Do<br />
inverse discrete wavelet transform after modifying the<br />
'<br />
'<br />
wavelet coefficients and get I i, j.<br />
All <strong>of</strong> I i, j are united<br />
' '<br />
to a watermarked image I = [ I ( i, j)]<br />
(1 ≤i, j ≤ N ).<br />
5) Define the fitness function using peak<br />
signal-to-noise ratio (PSNR) between I = [ I( i, j)]<br />
and<br />
' '<br />
I = [ I ( i, j)]<br />
.<br />
2 2<br />
N × max( I ( i, j))<br />
N N<br />
' 2<br />
PSNR = 10× log 10(<br />
)<br />
( Iij ( , ) − I( i, j))<br />
∑∑<br />
i= 1 j=<br />
1<br />
6) Create some random chromosomes into an original<br />
colony and give the values <strong>of</strong> s P and P m . Evaluate the<br />
fitness values <strong>of</strong> chromosomes and do the genetic<br />
operation until the process <strong>of</strong> GA stops and the optimal<br />
watermarked image I� is gotten.<br />
The final chromosome <strong>of</strong> GA and parameters<br />
( µ 1, µ 2, γ, x0, y0, P)<br />
<strong>of</strong> the two-dimensional chaotic<br />
Logistic map system can be looked upon the key <strong>of</strong> this<br />
watermarking scheme used in watermark extraction.<br />
IV. WATERMARK IDENTIFICATION<br />
To an encrypted meaningful watermark extracted from<br />
watermarked image, people maybe not distinguish its<br />
decrypted form through technical indexes, such as BCR,<br />
NC, and eyes, because <strong>of</strong> some interferential causations<br />
to watermarked image in the communication and usage.<br />
The SNN can effectively identify the extracted watermark<br />
in our former research[15] so that it is used in this scheme<br />
too.<br />
A. Synergetic Neural Networks<br />
The SNN model is a top-down network constructed by<br />
synergetic different from traditional network constructed<br />
by the method researched in single neuron’s characteristic,<br />
configuration and connection[16].<br />
© 2011 ACADEMY PUBLISHER<br />
(4)<br />
Dynamical system can be described by state vector in<br />
Synergetic. Let a state vector be q= ( q1, q2, � , q ' ) .<br />
M<br />
A synergetic associative pattern recognition system can<br />
be described by dynamical evolutionary process, in which<br />
the system evolves by neural network learning to fill<br />
incomplete data set and form pattern. Furthermore, let<br />
'<br />
prototype pattern number be M and prototype pattern<br />
'<br />
' '<br />
vector’s dimension be N , where satisfies M ≤ N .<br />
A dynamical equation can be described by (5).<br />
.<br />
∑ ∑ (5)<br />
q= λ v ( v q) −B ( v q) ( v qv ) −Cq<br />
( qq )<br />
+ + 2 + +<br />
k<br />
k k k<br />
'<br />
k≠k '<br />
k k k<br />
where q as recognizable pattern vector with original<br />
input value q0 = q(0)<br />
can be decomposed into<br />
prototype v k and remnant vector w , having<br />
'<br />
M<br />
∑ ξk<br />
k = 1<br />
k and<br />
+<br />
w = 0<br />
q= v + w<br />
vk . Attention parameter<br />
λ k is positive. B and C are appointed coefficients<br />
and must be more than zero. Prototype pattern vector k v<br />
is expressed as vk = ( vk1, vk2, � v ' ) ′ and<br />
kN<br />
+<br />
v k is an<br />
adjoint vector <strong>of</strong> v k , which satisfies an orthogonal<br />
⎧ ′<br />
+<br />
1,<br />
k = k<br />
condition vk vk<br />
= δ kk′<br />
= ⎨ . All v k will be<br />
⎩0,<br />
k ≠ k′<br />
normalized as<br />
'<br />
N<br />
∑ vkl<br />
= 0 and centered as<br />
l = 1<br />
'<br />
N<br />
2 1/2<br />
k = ( ) 1<br />
2 ∑ kl =<br />
l=<br />
1<br />
v v<br />
.<br />
Order parameter ξ k is defined as k k vq ξ<br />
+<br />
= . The<br />
dynamical equation can be rewritten by order parameter.<br />
k = k k − B∑ 2<br />
k′ '<br />
M<br />
k − C ( ∑<br />
2<br />
k′ ) k<br />
k′ ≠ k k′<br />
= 1<br />
ξ� λξ ξ ξ ξ ξ (6)<br />
D= ( B+ C) ∑ ξ , (6) is simplified to (7).<br />
k<br />
2<br />
Used '<br />
'<br />
k<br />
� (7)<br />
ξ ξ λ ξ<br />
2<br />
k = k( − D+ B k )<br />
So the SNN model is constructed with three layers.<br />
The top layer is the input layer. All order parameter<br />
neurons form the middle layer. The down layer is the<br />
output layer.<br />
B. Watermark Extracting<br />
The watermark extracting is the contrary producer <strong>of</strong><br />
watermark embedding. The DWT transforms <strong>of</strong> the<br />
received watermarked image I � and host image I<br />
could be done according to the rule <strong>of</strong> watermark<br />
embedding and the final chromosome <strong>of</strong> GA need be