IV. Experimental Results and Discussion
4.6 Tamper Detection
To evaluate the validity of the proposed image authentication algorithm and make up tampered images, we use Adobe Photoshop CS2 for implement of image processing operations. In our experiments, we let parameters q1=30, q2=10, K1=1234, K2=1234, L=1, β=3. Fig. 27~30 (a), (b), and (c) demonstrate the dual watermarked images (visible and semi-fragile watermark embedded), tampered images, and tampering detection images respectively. In Fig. 27 (b), one object (A .com logo) is inserted into the dual watermarked L
the vi dual
watermarked Baboon image. In the top rig rt of the watermark (logo) image, we use neighboring pixels to remove the visible watermark. In Fig. 29 (b), two objects (A .com logo and boat) are inserted into the dual watermarked Lake image. In the top part of the watermark (logo) image, we use neighboring pixels to remove the visible watermark. In Fig.
30 (b), three object (A .com logo and two Peppers) are inserted into the dual watermarked Peppers image. From the detection result of tampered images, the marked points indicate the tampered parts of watermarked image and these parts are located correctly.
ena image. In the shoulder part of the Lena image, we use neighboring pixels to remove sible watermark. In Fig. 28 (b), one object (A .com logo) is inserted into the
ht pa
(a) (b) (c)
Fig. 27 (a) Result (watermarked) image (b) Tampered image (c) Tampering detection
(a)
Fig. 28 (a) Result (waterma
(b)
rked) image (b) Tampered im
(c)
age (c) Tampering detection
(a)
Fig. 29 (a) Result (waterma
(b)
rked) image (b) Tampered im
(c)
age (c) Tampering detection
Fig. 30 (a) Result (watermarked) image (b) Tampered image (c) Tampering detection
2
(a) (b) (c)
For the combination of tampering operations and mild modifications, Fig. 31 and 32 show the tamper detection after AWGN with different σ and JPEG compression with different quality factor (QF). Fig. 31 (a) shows the dual watermarked Lake image. Fig 31 (b), shows the tampered Lake image: one object (a boat) is inserted into the dual watermarked Lake image. From Fig 31 (c) (d) (e), we can see the detection result of tampered Lake image is located correctly after AWGN with differentσ2. From Fig 32 (a) (b)
(c) (d), we can see the detection result of tampered Lake image is located correctly after JPEG compression with different quality factor (QF) setting.
From more serious attacks like watermark removal, we are also interested in the detection capability by the proposed approach. Fig. 33~36 demonstrates the tamper etection result. We can clearly see the tampered area are labeled and reflected the evidence f tampering.
Out of the above experiment results, we employ the normalized cross-correlation (NC) to evaluate the performance of watermark detection without post-processing (PP) operation.
The value of NC is calculated as formula (16). Table 5~13 shows the NC value after AWGN with different
d o
σ2 and JPEG compression with different quality factor (QF). From the
results from Fig. 33~36 and e not only
det G
ompression.
Table 5~13, we can see the authentication schem
ect tamper correctly but also tolerate mild modifications like AWGN and JPE c
(a) (b)
(c) (d) (e)
(f) (g) (h) F
im
ig. 31 Tamper detection for mixing tampering operations and AWGN (a) watermarked age (b) tampered image (c) σ2=6(d) σ2 =12 (e) σ2 =18 (f) σ2=24 (g) σ2 =30
(h) σ2=36.
(a) (b) (c)
(d)
Fig. 32 Tamper detection for m
(e) (f)
ixing tampering operations and JPEG compression (a) QF=100 (b) QF=90 (c) QF=80 (d) QF=70 (e) QF=60 (f) QF=50.
(a) (b) (c)
Fig. 33 (a) Dual watermarked image of Lena (b) Tampered dual watermarked image with watermark removal attack (c) Tampering Detection
c)
Fig. 34 (a) Dual watermarked image of Baboon (b) Tampered dual watermarked image with watermark removal attack (c) Tampering Detection
(a) (b) (
(a) (b) (c)
Fig. 35 (a) Dual watermarked image of Lake (b) Tampered dual watermarked image with watermark removal attack (c) Tampering Detection
(a) (b) (c)
Fig. 36 (a) Dual watermarked image of Peppers (b) Tampered dual watermarked image with watermark removal attack (c) Tampering Detection
' ''
Table 5 Lena (NCTU logo): Robustness against AWGN and JPEG compression.
NC =
Table 6 Lena (IIM logo): Robustness against AWGN and JPEG compression.
Image Lena (IIM logo)
Table 7 Baboon (NCTU logo): Robustness against AWGN and JPEG compression.
Image Baboon (NCTU logo)
Table 8 Baboon (IIM logo): Robustness against AWGN and JPEG compression.
Image Baboon (IIM logo)
AWGN:σ2 6 12 18 24 30 36
NC 0.94 0.85 0.77 0.71 0.67 0.63
JPEG:QF 100 90 80 70 60 50
NC 0.99 0.98 0.92 0.82 0.72 0.65
Table 9 lake (NCTU logo): Robustness against AWGN and JPEG compression.
Image lake (NCTU logo)
6 12 18 24 30 36 AWGN:σ2
NC 0.94 0.85 0.77 0.72 0.67 0.63
JPEG:QF 100 90 80 70 60 50
NC 0.99 0.98 0.93 0.85 0.75 0.67
Table 10 lake (IIM logo): Robustness against AWGN and JPEG compression.
Image lake (IIM logo)
able 11 Peppers (NCTU logo): Robustness against AWGN and JPEG compression.
T
Table 12 Peppers (IIM logo): Robustness against AWGN and JPEG compression.
Image Peppers (IIM logo)
AWGN:σ2 6 12 18 24 30 36
NC 0.93 0.85 0.77 0.71 0.66 0.63
JPEG:QF 100 90 80 70 60 50
NC 0.99 0.98 0.93 0.86 0.77 0.70
In our dual watermarking algorithm, we apply Bi18/10 filter for semi-fragile watermark lter from the ex tal results. Fig. 37 (a) (b) (c) on after JPEG compression with diff quality r by u i18/10 . Fig. 37 (d) (e) (f) is the tamper detection after
J ression with diffe quality r by us fi 8 ( ) (c) is the
mper detection after AWGN with different
not Bi9/7 filter. Because we find Bi18/10 have better robustness than Bi9/7 fi
perimen is the tamper detecti
erent facto sing B filter
PEG comp rent facto ing Bi9/7 lter. Fig. 3 a) (b
ta σ2 by using Bi18/10 filter. Fig. 38 (d) (e) (f) is
the tamper detection after AWGN with different σ2 by using Bi9/7 filter. In advance, Table ows the C value these GN with different
14~15 sh N from filters after AW σ2 and JPEG
compression with different quality factor (QF). It is clear that semi-fragile watermark
algorithm by using Bi18/10 filter have more robust than using Bi9/7 filter.
(a) (b) (c)
(d e) (f)
Fig amper tion af G com ion using different filters (QF) (a) QF=80 by u /10 F b) QF=70 by using Bi 18/10 Filt ) QF=60 by using Bi 18/10 Filter (d) QF=80 by using Bi 9/7 (e) QF y usin /7 Filte F=60 ing Bi
9/7 Filter
) (
. 37 T Detec ter JPE press
sing Bi 18 ilter ( er (c
Filter =70 b g Bi 9 r (f) Q by us
(a) (b) (c)
(d) (e) (f)
Fig. 38 Tamper Detection after AWGN using different filters (a) σ2 =6 by using Bi 18/10 Filter (b) σ2 =12 by using Bi 18/10 Filter (c) σ2=18 by using Bi 18/10 Filter (d)
2 6 2
σ = by using Bi 9/7 Filter (e) σ =12 by using Bi 9/7 Filter (f) 2= 9/7 Filter
Table 13 Lena (NCTU logo): Robustness against AWGN and JPEG compression by using Bi18/10 Filter.
σ 18 by using Bi
Image Lena (NCTU logo)
AWGN:σ2 6 12 18 24 30 36
NC 0.94 0.85 0.78 0.71 0.67 0.63
JPEG:QF 100 90 80 70 60 50
NC 0.99 0.99 0.93 0.86 0.76 0.69
Table 14 Lena (NCTU logo): Robustness against AWGN and JPEG compression by using i9/7 Filter.
B
Image Lena (NCTU logo)
6 12 18 24 30 36 AWGN:σ2
NC 0.90 0.77 0.68 0.62 0.59 0.56
JPEG:QF 100 90 80 70 60 50
NC 0.99 0.96 0.85 0.75 0.67 0.62
For general applying our algorithm, we use scrambling technique to generate binary watermark and embed it into rectangle image as semi-fragile watermark. Fig. 39 (a) shows the dual watermarked image of Lena and the size is 512×480. Fig. 39 (b) shows tampered dual watermarked image and the size is 512×480. We add one eye on the hat in the tampered image.
From the detection result of tampered image, the marked points indicate the tampered parts of watermarked image and these parts are located correctly.
Fig. 39 (a) Dual watermarked image of Lena and the size is 512×480 (b) Tampered dual watermarked image and the size is 512×480 (c) Tampering Detection
After the intensive perform nce comparison, the results of different attacks, visual quality
analyses and te e color image
watermarking by using dual watermarks with HVS method is more robust with better image quality. In summary, we are convinced that the proposed complete architecture is a superior scheme among the referred published techniques.
a
mper detection demonstrate that the proposed multipurpos