Information Technology Reference
In-Depth Information
even rose with increase of
S
in the case of Lena, meaning that the image was
“repaired” somewhat by dithering.
Baboon and Lena QIM-embedded in the dual-transform domain were also tested in
the experiments. Watermarks were embedded in the 50th double-layered transform
coefficients with ∆=60 in both images. The results are given in Figures 10 and 11.
Similar to the previous experiments, the embedded data became undetectable as
S
approached and exceeded
. PSNR showed an increase when
S
was not too large, and
then went down. In both images, PSNR remained at a high level within the entire
range of the tested
S
values.
∆
0.6
45
0.4
40
0.2
35
0
20
40
60
80
100
20
40
60
80
100
S
S
(a) Relative bit error rate vs.
S
(b) PSNR vs.
S
Fig. 10.
Removal of dual-transform/QIM watermark in Baboon
0.6
50
0.4
45
0.2
40
0
20
40
60
80
100
20
40
60
80
100
S
S
(a) Relative bit error rate vs.
S
(b) PSNR vs.
S
Fig. 11.
Removal of dual-transform/QIM watermark in Lena
6
Conclusions
It has been demonstrated that the presence of QIM-embedded information can be
detected based on analyses of coefficient histograms. The QIM embedding is also
vulnerable to active-warden attacks as extraction of the embedded data can be pre-
vented by a dithering attack on the coefficients. Therefore, unless taking special
measures (currently under investigation), QIM embedding is not recommended for
