Generating word images using deep generative adversarial networks

Abstract

As one of the most important research topic of nowadays, deep learning attracts researchers' attention with applications of convolutional (CNNs) and recurrent neural networks (RNNs). By pioneers of the deep learning community, generative adversarial training, which has been working for especially last two years, is defined as the most exciting topic of computer vision for the last 10 years. With the influence of these views, a new training approach is proposed to combine generative adversarial network (GAN) architecture with a cascading training. Using CVL database, text images can be generated in a short training time as a different application from the existing GAN examples.