Cryptography Reference
In-Depth Information
one bit may make the extracted value very different from the embedded one.
To overcome this problem, we embed each floating value multiple times into
the image. During the extraction, we determine the extracted bits in the ma-
jority cases. This process can be shown in Fig. 11.21. A further problem is
how to guarantee the security of the watermarks. This is especially important
for the copyright watermark. In our system, we use the Interleaving Embed-
ding Technique. We randomly select the columns of DCT blocks to embed
one of the watermarks, so perhaps the first column is used to embed the fea-
ture watermark. The second column may be used to embed the annotation
watermark, and so on. The selection scheme can be viewed as an embedding
key.
Feature
watermark
Copyright
watermark
Watermark to
be embedded
Fig. 11.21.
The embedding strategy.
11.4.3 Online Image Retrieval with Various Query Strategies
After embedding the watermarks into all the images in the database, we can
now do the online retrieval. Here we can use several kinds of queries. The most
normal is feature-based. We can also use copyright-based and annotation-
based, or alternatively we can use the combination query schemes.
For the feature-based query scheme, we first extract the features from the
query image. Note that if the query image is without a feature watermark, we
should compute the feature online for it. Then we compare the query feature
vector with the feature watermark in each database image to find the most
desired similar images. For the feature-and-annotation-based query scheme,
we can only search the desired images with the same annotation. In general,
it is the semantic of the image. For the feature-and-copyright-based query
scheme, we can only search the desired images having the same copyright. For
