Coloring Ancient Egyptian Paintings with Conditional Generative Adversarial Networks

Abstract

The aim of colorizing gray-scale images is to turn a gray-scale image into a real-looking color image, which is still a difficult task. In this paper, we present a new fully automated colorization technique to assign realistic color images with high levels of textured details, with fewer time and storage requirements than the most recent techniques. Our presented model is designed as a Conditional Generative Adversarial Network with a generator and discriminator to colorize Ancient Egyptian Paintings. Our model is trained using a novel dataset that is aggregated from Ancient Egyptian Paintings and contains more than 1000 images. Our model and traditional deep neural networks are assessed using Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), and Mean Square Error (MSE). The outcomes demonstrate the presented technique's ability to colorize images realistically and naturally while attaining state-of-the-art results.