Master’s Thesis Proposes an Advanced Data Hiding System Using Deep Learning

Duhaa Fadill Abbas
The Department of Software at the College of Information Technology, University of Babylon, held the defense of the Master’s thesis submitted by Mohammed Majed Jabbar, entitled “Proposing a Data Hiding System Based on Deep Learning Techniques and Texture Context.” The defense was attended by the examination committee, faculty members, and researchers, and was conducted in a rigorous academic atmosphere reflecting the advanced scientific level of the research and its significance at the intersection of information security, digital image processing, and deep learning technologies. The study aligns with contemporary trends in developing intelligent solutions aimed at strengthening digital data security.

The thesis highlighted the challenges associated with conventional image steganography techniques, which often suffer from limitations in robustness, security, and imperceptibility. Moreover, such traditional approaches are vulnerable to statistical and geometric attacks due to their reliance on fixed embedding regions without adequately considering the structural, geometric, and textural characteristics of the cover image. The study emphasized that these shortcomings negatively affect both the security of the concealed message and the quality of the cover image, thereby underscoring the need for a more intelligent and adaptive methodology that accounts for the intrinsic properties of the image prior to embedding the secret data.

To address these limitations, the researcher proposed an advanced data hiding framework that integrates deep learning techniques with texture context and spatial structural analysis to enhance security and imperceptibility during both embedding and extraction processes. The proposed system employs the MiDaS model to generate a depth map of the cover image, enabling a comprehensive understanding of structural relationships within the image. This is followed by the application of the Shi–Tomasi algorithm to extract key points, along with a proposed point-merging mechanism designed to eliminate redundancy and reduce computational complexity.

Furthermore, an irregular triangular mesh is constructed using the Delaunay Triangulation algorithm, after which the correlation degree among pixels within each triangle is calculated using the Spearman correlation coefficient. Based on the computed correlation values, the embedding strategy is dynamically adapted according to categorized correlation levels, thereby enhancing both embedding efficiency and security.

This study represents a valuable scientific contribution to the field of data hiding and information security. It reinforces the integration of deep learning methodologies in the development of more intelligent and efficient protection systems, thereby supporting ongoing research efforts and keeping pace with rapid advancements in cybersecurity and digital image processing technologies.