Tile Art Image Generation Using Conditional Generative Adversarial Networks

Abstract

Image-to-image translation is a task of mapping an image in one domain to a corresponding image in another domain. The task includes various types of problems such as super-resolution, colorization, and artistic style transfer. In recent years, with the advent of deep learning, the technology has been rapidly advanced. The main purpose of this paper is to propose a tile art image generation method using machine learning approach based on conditional generative adversarial networks. To make the training data set of tile art images, we adopted a square-pointillism image generation method using the greedy approach. After training, the proposed network can generate tile art images that have the structure of tiles and reproduce the original images well. As regards generating time, the greedy approach takes 1322 seconds to generate tile art image of size 4096x3072, while the proposed machine learning approach takes 0.593 seconds.