Fake Surface Generation Based on GAN (Generative Adversarial Network) and Mesh Structure

The image processing method based on geometric information in graphs combines principles from graph theory, geometry, and data science. Using topological and geometric concepts, artificial graphs can be generated to resemble real data while preserving the original graphs' fundamental properties and structural patterns. These artificial graphs can then be used to construct raster images from vector images. Creating a synthetic Earth surface image based on landmarks extracted from real satellite images is a significant challenge. The proposed system introduces a new approach for generating fake Earth surface images derived from real satellite imagery. This hybrid method consists of three main stages: converting satellite images into irregular triangular mesh representations, using a modified Generative Adversarial Network (GAN) with an added fully connected layer in both the generator and discriminator networks to produce synthetic mesh images, its name One Fully Connected GAN (OFCGAN) and constructing raster images from vector data. The system achieved an FID rate of 230.83, demonstrating its effectiveness in generating realistic raster images. The study aims to develop raster images that can be utilized in applications such as information warfare and to produce graphs that are simulated to the original graphs. The method ensures realistic image replication while maintaining visual and structural integrity.