| dc.description.abstract | In the last few years the use of digital image has a grown tremendously at the same time tools
to manipulate digital multimedia are also increasing such as Paint, Photoshop, Corel Draw, etc.,
and became very easy to use. To ensure trustworthiness, image authentication techniques, such as
image hashing, watermarking have emerged to verify content integrity and prevent forgery.
Traditionally data integrity issues are addressed by cryptographic hashes or message authentication
functions, which are key-dependent and sensitive to every bit of the input message. As a result,
the message integrity can be validated when every bit of the message is unchanged. The definition
of authenticity for multimedia is not as straightforward. So this type of hashing cannot be use
for authenticate content because an image hash function takes into account changes in the visual
domain. In particular, a perceptual image hash function should have the property that two images
that look the same to the human eye map to the same hash value, even if the images have different
digital representations. Therefore bit-by-bit verification is no longer a suitable way to authenticate
multimedia data and a media authentication tool that validates the content is more desired. An
immediately obvious application for a perceptual image hash is identification/search of images in
large databases. Several other applications have been identified recently in content authentication,
watermarking, and anti-piracy search.
This work introduces the use of image normalization algorithm and SVD decomposition Hash
function to generate the Hash vector for digital images. SVD decomposition is an important linear
algebra tool, which is often used in pattern recognition, digital watermarking and other signal
processing fields, on the other hand image normalization is an algorithm that has been shown to
be robust to different kinds of geometric attacks such like, rotation, scaling and flipping. Using the
image normalization algorithm as preprocessing step we can increase the robustness of the hash
value as it shown by numerical results.
To measure the performance of image hashing, we choose the Hamming distance between the
binary hashes, normalized with respect to the length (L) of the hash as a performance metric. This
is expected to be close to 0 for similar images and close to 1 for dissimilar ones. As more parts of a
picture are changed, the manipulated image and the original image become more dissimilar. This
work proposes by numerical results the use of 0,1 in terms of Hamming distance as a threshold. | |
