Cryptography Reference
In-Depth Information
prove that NHS∈[0, 1]. If we acquire the higher NHS values, the embedded
watermark is more similar to the extracted one. Fig. 11.7 shows the robust
and semi-fragile watermarks extracted from the watermarked image which
has not been subjected to attack. Both NHS values are equal to 1.0. This
means that the proposed algorithm has been able to extract the watermarks
perfectly. That is because the embedded watermarks and the extracted ones
are identical. We can see that, in the proposed method, the robust watermark
is embedded first and does not modify the VQ compressed cover work with
the first stage codebook. The robust watermark doesnt affect the semi-fragile
part. We do not use the two stage codebooks but we do use the product
codebook in the extraction process. The semi-fragile and robust watermarks
are independent as shown in Fig. 11.7.
(a) (b)
Fig. 11.7.
Extracted watermarks under no attack. (a) Extracted robust watermark,
NHS = 1.0. (b) Extracted semi-fragile watermark, NHS = 1.0.
To check the robustness and fragility of our algorithm, we have made
several attacks on the watermarked image. These include JPEG compression,
VQ compression, spatial image processing and rotation. We also do the exper-
iment of watermark extraction from the original image in which no watermark
is embedded. We now give the experimental results in five subsections.
Watermark Extraction from the Original Image
The experimental results of watermark extraction from the original image are
shown in Fig. 11.8. In this experiment, we use not the product codebook in
the extraction process but the two-stage VQ codebooks. From the results, it
is clear that, although we can extract the robust watermark using NHS =
1.0 from the original image, we cannot extract the semi-fragile watermark
from the original image. For the first stage, each original image block can
be encoded using the same codeword as the watermarked compressed image
block with the first stage VQ codebook. Note that the robust watermarking
does not modify the compressed image and the watermarked image is the
same as the VQ compressed image in the first stage. The index properties of
the original image are the same as that of the compressed image using the
first stage VQ codebook. We can extract the exact robust watermark from
the original image. For the fragile part, the un-watermarked index obtained
is different from that of the watermarked compressed image of the second
