Cryptography Reference
In-Depth Information
Codebook
C
u
Inverse
Permutation
Codebook
C
key
4
Encoded
Input
Vectors
Normal VQ
Encoder
X
W
Division
Indices
Segmentation Position
key
5
Indices of Stage 1
Indices of Stage 2
Index
Segmentation
Index Polarities
Computation
Watermarking Position
key
3
Bit judgment
P
′
Extracted Bits
key
2
XOR
Extracted Semi-Fragile Watermark
W
EF
Composition
W
EPR
key
1
Inverse
Permutation
Extracted Robust Watermark
W
ER
Fig. 11.4.
The watermark extraction process.
11.2.3 Experimental Results
To evaluate the performance of the proposed method, the 512512 Lena image
with 8 bits/pixel resolution is used for multipurpose watermarking. The Lena
image is divided into 16384 blocks of size 44 for VQ encoding. A binary
image of size 3232 is replicated 16 times to obtain a binary watermark W
F
with size 128128 for semi-fragile watermarking. Another binary watermark
W
R
with size 128128 is used for robust watermarking. The original Lena
image and two watermarks are shown in Figs. 11.5(a)(c). The codebook C
1
with size 16 is used for Stage 1. The codebook C
2
of size 256 is used for
Stage 2 are obtained by use of the LBG algorithm [22]. This corresponds to
4 + 8 = 12 bits per overall index. If we embed the semi-fragile watermark
in the second stage, we can randomly select the watermarking position key
3
ranging from 0 to 7 for the semi-fragile watermarking. Before extraction, the
equivalent product codebook C of size 16256 = 4096 is generated by the
Cartesian product C
1
C
2
.
We first make an experiment upon the order of embedding using robust
and semi-fragile watermarks to show why we should embed the robust water-
mark in the first stage and the semi-fragile watermark in the second stage.
Fig. 11.6(a) shows the watermarked image with PSNR = 30.553 dB obtained
by use of the proposed method. Fig. 11.6(b) shows the watermarked image
with PSNR = 26.105 dB obtained by using the algorithm in the reverse em-
bedding order. We see that the proposed embedding order produces higher
