Synthesizing Handwritten Characters Using Naturalness Learning

In this paper we show how to synthesize handwritten characters using a proposed system for naturalness learning. We begin by explaining what we mean by naturalness and then show that in many characters, certain properties of font character strokes does not have a linear relation with this naturalness. This observation inspires the idea of using nonlinear techniques to model the naturalness in order to generate handwriting of a unique, personalized, form. Several techniques for achieving this were tested. Surprisingly, RNN with a recurrent output layer performed the best at generating characters very similar to a person´s handwriting.