248
Maria Calagna and Luigi V. Mancini
Fig. 11.4. The cover aerial picture is on the left, while the picture on the right represents
the watermarked aerial picture obtained by applying the SVD scheme with block size 8 and
watermark strength 0.3
schemes. The last column shows the quality difference between the attacked image
and the original cover when our scheme is applied. These experiments give evidence
that our watermarking scheme is robust against that kind of attacks, including the
high-distortive ones. Actually, it depends on the specific application if the level of
distortion is acceptable or not. We illustrate the worst case in Table 11.1 in order to
show the robustness properties of our scheme that is applicable in other domains that
could be different from the GIS field.
Regarding filtering techniques, we can see that the larger the window size is,
the more distortive the attack is. Regarding downsampling/upsampling attack, the
choice of the first pair of downsample/upsample factor leads to lower distortion,
while the choice of the second pair leads to higher distortion. Regarding JPEG compression and the addition of white Gaussian noise (AWGN) we considered the parameters that lead to more distortion in the attacked image: the quality factor is 10
in the JPEG compression, while the noise strength is 20 in the AWGN attack. Our
scheme is robust against contrast enhancement techniques that improve the visual
quality of the image and are considered high-frequency preserving [10]. They include histogram equalization and intensity adjustment as well. Finally, we show the
robustness results against clipping attack. To summarize, our scheme is more robust than the Cox scheme against median filtering, downsampling/upsampling, histogram equalization, intensity adjustment, and clipping attacks, while it is comparable to that scheme against some other attacks, including JPEG compression and
AWGN.
Regarding geometrical attacks, including rotation, translation, and scaling, our
claim is that our algorithm is robust against this kind of attacks, since the following
properties hold:
Maria Calagna and Luigi V. Mancini
Fig. 11.4. The cover aerial picture is on the left, while the picture on the right represents
the watermarked aerial picture obtained by applying the SVD scheme with block size 8 and
watermark strength 0.3
schemes. The last column shows the quality difference between the attacked image
and the original cover when our scheme is applied. These experiments give evidence
that our watermarking scheme is robust against that kind of attacks, including the
high-distortive ones. Actually, it depends on the specific application if the level of
distortion is acceptable or not. We illustrate the worst case in Table 11.1 in order to
show the robustness properties of our scheme that is applicable in other domains that
could be different from the GIS field.
Regarding filtering techniques, we can see that the larger the window size is,
the more distortive the attack is. Regarding downsampling/upsampling attack, the
choice of the first pair of downsample/upsample factor leads to lower distortion,
while the choice of the second pair leads to higher distortion. Regarding JPEG compression and the addition of white Gaussian noise (AWGN) we considered the parameters that lead to more distortion in the attacked image: the quality factor is 10
in the JPEG compression, while the noise strength is 20 in the AWGN attack. Our
scheme is robust against contrast enhancement techniques that improve the visual
quality of the image and are considered high-frequency preserving [10]. They include histogram equalization and intensity adjustment as well. Finally, we show the
robustness results against clipping attack. To summarize, our scheme is more robust than the Cox scheme against median filtering, downsampling/upsampling, histogram equalization, intensity adjustment, and clipping attacks, while it is comparable to that scheme against some other attacks, including JPEG compression and
AWGN.
Regarding geometrical attacks, including rotation, translation, and scaling, our
claim is that our algorithm is robust against this kind of attacks, since the following
properties hold: