Media of University of Babylon | Ph.D. Dissertation at the College of Information Technology Discusses “Generating deepfake surface images based on generative adversarial network

Media of University of Babylon - كلية الطب

Ph.D. Dissertation at the College of Information Technology Discusses “Generating deepfake surface images based on generative adversarial network

by : Duhaa Fadill Abbas

Date : 02/9/2025 30:15:08 PM

Veiw : 52

Ph.D. Dissertation at the College of Information Technology Discusses “Generating deepfake surface images based on generative adversarial network”

Duhaa Fadill Abbas
The Department of Software at the College of Information Technology held a Ph.D. dissertation defense titled “Generating deepfake surface images based on generative adversarial network” by doctoral candidate Raed Abdul-Ridha Sheikhan, under the supervision of Prof. Dr. Tawfiq Abdul-Khaliq Abbas. The defense took place at 9:00 a.m. on Tuesday, September 2, 2025, in the College’s main conference hall.

The dissertation emphasized the significance of image generation techniques in scientific fields such as remote sensing, simulation, and geoinformatics, highlighting the remarkable advancements achieved through deep learning. Among these, Generative Adversarial Networks (GANs) have emerged as one of the most effective approaches for producing high-quality synthetic images. GANs are trained on real data distributions, enabling the generation of highly realistic synthetic images that closely resemble genuine ones.

To define the scope of the study, the researcher compared the proposed work with prior studies on GANs, which primarily focused on generating raster images by training GANs directly on pixel data. While effective, this pixel-based approach poses challenges due to the computational complexity of raster data and the enormous resources required to process it.

In contrast, this dissertation focuses on the geometric representation of raster images and introduces a novel four-stage framework.
Stage One involves preprocessing satellite raster images of the Earth’s surface, including grayscale conversion, image enhancement, and geometric feature extraction using the Scale-Invariant Feature Transform (SIFT) algorithm. A mesh structure is then constructed using Delaunay triangulation based on the extracted features, transforming raster images into mesh-based geometric representations that capture the topological and spatial properties of satellite imagery.

Stage Two proposes an enhanced GAN model by integrating a fully connected layer after both the generator and discriminator within the baseline GAN architecture. This new model, termed the One Fully Connected GAN (OFCGAN), processes low-dimensional vector inputs (X and Y coordinates) with minimal computational complexity. The aim is to generate synthetic mesh structures that preserve the geometric integrity and distribution patterns of real surface data during training.

Stage Three evaluates the proposed network using test data to validate its performance.

Stage Four reconstructs the raster image by transferring texture information from real raster imagery onto the synthetic mesh structure, producing a synthetic raster image that closely mimics the real Earth surface image.

This research contributes to advancing deep learning-based image generation, offering a more efficient and geometry-oriented approach for producing high-fidelity synthetic images in applications involving remote sensing and geospatial analysis.

ت	اسماء اعضاء لجنة المناقشة	اللقب العلمي	الاختصاص الدقيق	مكان العمل	المنصب
1	د. حسين كيطان منسي	استاذ	تعدين بيانات و ذكاء اصطناعي	كلية الرافدين الجامعة / هندسة تقنيات الحاسوب	رئيساً
2	د. هضاب حالد عبيس	استاذ	ذكاء اصطناعي	جامعة بابل	عضوا
3	د. الحارث عبدالكريم عبدالله	استاذ	امن معلومات و شبكات	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا
4	د. امنة عطية داوود	استاذ مساعد	ذكاء اصطناعي	رئاسة جامعة بابل / مركز الحاسبة الالكترونية	عضوا
5	د. اعتماد رحيم علي	استاذ مساعد	وسائط متعددة	جامعة تكنولوجيا المعلومات و الاتصالات / كلية معلوماتية الاعمال	عضوا
6	د. توفيق عبد الخالق عباس	استاذ	معالجة صور	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا و مشرفا