Machine Learning in Image Authentication

Citation:
Bakrawy, L. M. E., A. - E. " "Hassanien, and N. I. Ghali, Machine Learning in Image Authentication, , Cairo, Al-Azhar University, 2012.

Thesis Type:

PhD thesis

Abstract:

In recent years image authentication has gained substantial attraction by the research community. It promises the solution to many problems such as content piracy, illicit manipulation of medical/legal documents, content security and so on. As watermark-based image authentication approaches are efficient and attractive, some types of watermarks such as logos, labels, trademark, or random sequence representing the author’s ownership, are mbedded into the desired digital image. Generally, a registration to the authentication center is necessary, which helps to solve ownership disputes by identifying the owner of the disputed media. If necessary, the embedded watermark in the digital image can be used to verify ownership Due to the open environment of Internet downloading, copyright protection introduces a new set of challenging problems regarding security and illegal distribution of privately owned images. One solution to these problems is digital watermarking, i.e., the insertion of information into the image data in such a way that the added information is not visible and yet resistant to image alterations. A watermarking technique is to prevent digital images that belong to rightful owners from being illegally commercialized or used, and it can verify the intellectual property right. The watermark should be robust and transparent, but the ways of pursuing transparency and robustness are conflict. For instance, if we would like to concentrate on the transparency issue, it is natural to embed the smallest modulation into images whenever possible. However, due to such small values in the embedded watermark, attacks can easily destroy the problems The first proposed solution is based on the associative watermarking and vector quantization. It achieves more effective against several images processing such as blurring, sharpening adding in Gaussian noise, cropping, and JPEG lossy compression especially in case of Gaussian noise and blurring. Also this technique is implemented to hide biometric data, fingerprint image, over three different types of medical images: CT, MRI and interventional images. It also achieves an effective resistance against several images processing such as JPEG lossy compression, sharpening, blurring and adding in Gaussian noise The second contribution in this thesis is strict authentication of multimodal biometric images using an improved secure hash function (ISHA-1) and near sets. It indicates that the proposed hash function is collision resistant and assures a good compression and preimage resistance. Also it reduces the time of implementation comparing to standard secure hash function. Moreover, the difference in time between SHA-1 and ISHA-1 increases by increasing the number of letters in message since the running time of implementation of ISHA-1 is limited compared to the running time of implementation of SHA-1. Also the proposed approach guarantees the security assurance and reduces the time of implementation. The third proposed contribution is fragile watermarking approach for image authentication based on rough k-means only and hybridization of rough k-means and particle swarm optimization. It can embed watermark without causing noticeable visual artifacts, and does not only achieve superior tamper detection in images accurately, it also recovers tampered regions effectively. In addition, it shows that the proposed approach can effectively thwart different attacks, such as the cut-and paste attack and collage attack, while sustaining superior tamper detection and localization accuracy. Moreover, the running time of implemented hybrid system is limited compared to the running time of the implemented rough k-means only. Especially, when we used exponential particle swarm optimization to optimize the parameters of rough k-means.

Tourism