11, 12, 13] has also been proposed as an alternative image authentication
technique. Both digital watermarking and authentication signature are con-
sidered active image authentication techniques. They respectively requires
a known signal to be embedded into an image or the content features to
be extracted from an image before the image can be authenticated. Re-
cently, a passive-blind image authentication approach was proposed [14, 15].
The passive-blind image authentication approach does not require any prior
knowledge from an image for content authentication and tampering detec-
tion. The passive-blind technique would be useful in the situation where the
opportunity for embedding an active authentication signal on an image does
not present itself, as often the case for various image forensics situations.
These techniques are important for application such as criminal investiga-
tion, trustworthy journalistic reporting, and intelligence analysis.
We expect to see a plethora of proposed new techniques for tackling
the related open issues in the passive-blind image authentic research. To
assess the merits of the proposed techniques by various researchers, there
should be a way to measure how well the technique has solved its intended
problem. In situation where there is a lack of a good mathematical model
for the authentication object such as fake images, an empirical model can
be realized through a dataset, e.g., an image-splicing detection algorithm
can be evaluated on a dataset with spliced images. On the other hand,
the performance of a technique evaluated on a specific dataset may be very
different from the results obtained using another different dataset, due to
the possible bias within the different datasets. This points to the importance
of proper dataset design and the need for having a common dataset for a
fair comparison of various proposed techniques. The availability of such a
common and proper dataset would help to expedite the progress of a thriving
research area.
There are a number of image dataset available for various types of im-
age processing research; content-based image retrieval community commonly
uses Corel image dataset, digital watermarking community has a dataset put
together by Fabien Petitcolas [16], face recognition research has Yale face
database [17] and other general image processing research can use USC-SIPI
Image Database [18]. The issue is whether we can reuse one of the available
image dataset for the research of passive-blind image authentication. Eval-
uation of the passive-blind image authentication techniques requires fake
images as well as authentic images with a reliable origin, and these im-
ages, particularly the fake images, are not readily available. Therefore, an
effort is needed for collecting suitable datasets for the passive-blind image
authentication research. For instance, the problem of image splicing [15] is
2