Style-driven Handwritten Character Generation

Abstract

Handwritten character generation is a popular topic with a variety of applications. Many methods of character generation have been proposed in the literature, but few of these methods focus on preserving the writer's style while producing handwritten characters. Representing handwriting styles involves the challenge of representing both the style of each character and the overall style of the writer. In this study, unlike the studies in the literature, it is tried to produce the characters of the person that we have not seen yet by extracting the handwriting style of the person with limited data. While producing handwritten characters, spline curves of the characters were used as input, as well as handwritten images. In the developed method, a multi-tasking learning network is proposed by using the Conditional Variable Autoencoder (CVAE) model, which is one of the deep learning methods. In the experiments carried out, three different models were compared and the performance of the Conditionally Variable Autoencoder (CVAE) network trained with spline curves gave better results compared to other models.