Image Synthesis using U-net: Sketch 2 Image

Abstract

Image synthesis has been an important part of digital art, fashion design, and law enforcement, among others. In this paper, we introduce Sketch2Image, an automatic system for converting hand-drawn sketches to realistic images based on Conditional Generative Adversarial Networks (cGANs). The model employs a U-Net-based encoder and decoder to produce high-quality images with detailed finesse. The feasibility study takes into consideration technical, operational, economic, and scheduling factors, guaranteeing practicability and effectiveness. The incremental development approach is followed in the project, guaranteeing iterative improvement and performance boost. The assessment is done based on metrics like Mean Squared Error (MSE), Structural Similarity Index (SSIM), and adversarial loss, guaranteeing model efficacy. Experimental results confirm the system's capacity for creating visually realistic and contextually relevant images, with potential applications in creative and investigative fields.