03/08/2024
By Danielle Fretwell

The Francis College of Engineering, Department of Electrical and Computer Engineering, invites you to attend a Doctoral Dissertation defense by Othmane Habbouli on: A Blind, High Capacity, Robust to Noise, Automatic Secure Self Recovery Information Hiding/Authentication.

Candidate Name: Othmane Habbouli
Degree: Doctoral
Defense Date: March 20, 2024
Time: 11 a.m. (EST)
Location: This will be virtual defense via Zoom. Those interested in attending should contact committee chair Dalila_Megherbi@uml.edu.

Committee:

  • Advisor: Professor Dalila B. Megherbi, Faculty ECE Department, Director CMINDS Center, Ph.D.
  • Professor Xuejun Lu, Faculty ECE Department, UML, Ph.D.
  • Professor Kanti Prasad, Faculty Emeritus, ECE Department, UML, Ph.D.
  • Kathryn Millis, Program Lead Manager, US Air Force, Hanscom Base, Ph.D.

Brief Abstract:

The high demand for securely exchanging digital images with secret hidden information over the internet has become imperative as technology has evolved over the past years. It is almost impossible to avoid unpredicted attacks such as cropping and unwanted added noise to images caused by digital circuits or unauthorized hackers during transmission. Additionally, the need to send a maximum number of hidden embedded images per carrier has always been a challenge for researchers to overcome, given the limitation of low capacity or load in the embedding information while keeping transparency in the transmitted images with hidden information. This thesis proposes a blind, high-capacity, robust-to-noise, robust-to-encryption, JPEG compression, high security, automatic tampering detection, and self-recovery information-hiding scheme. The proposed blind method, blind in the sense that a carrier image is unavailable at the destination, uses different block sizes, 2x2, 4x4, and 8x8, to obtain various results. It can embed and hide sub-image DCT moments of several full gray-scale hidden images (as opposed to binary) and several full gray-scale watermarking images, each of the exact full sizes as a given arbitrary carrier host image, into the intensities of the host carrier image with very high imperceptibility, instead of hiding information in the moments of the carrier image, as is the case with existing classical schemes, in general. Two primary goals have been accomplished in this study. First, we show how the proposed blind algorithm allows a maximum number of grey images to be embedded into a carrier image. Second, we show how our scheme is resilient and effective against unpredicted noise (salt and pepper) up to a certain degree level applied by a foreign source. Unlike other classical schemes, where DCT moments are embedded in the moments of the carrier image, our proposed DCT-based scheme proved to be noise-resilient and outperformed the classical scheme. We also proved this concept by modeling and mathematically analyzing the impact of noise on the extracted reconstructed hidden image using the proposed scheme (DCT into intensities) and its superior performance and comparison results to classical schemes (DCT into DCT). Other contributions include self-recovery capability to recover as much as possible (if not most information) in case of unauthorized cropping attacks from hackers, resiliency against encryption, and JPEG Compression. To do that, we tackle the challenging case usage of redundancy in the hidden information while keeping the resulting watermarked image transparency. As a result, we can, via a blind scheme, hide eight arbitrary hidden gray-scale images with redundant information into a random carrier image. This uses two factors, as in this group’s prior research work, to make the carrier image independent of the secret images and ensure that an unauthorized third party will not easily tamper with the extracted images. Our analysis illustrates that DCTs are more compact and have higher reconstruction accuracy than LMs (Legendre Moments) and TMs (Tchebichef Moments).