Probabilistic synthesis of personal style handwriting

Abstract

The goal of personal style handwriting synthesis is to produce texts in the same style as an individual writer by analyzing the writer`s handwriting samples. The difficulty of handwriting synthesis is that the output should have the characteristics of the person`s handwritings as well as looking natural, based on a limited number of available examples. We represent the former as strokes for shape, position/size of character component for layout, and linear transformation for global appearance. For the latter, we specialize the character component production depending on preceding character component. Such specialization is modeled by a probabilistic distribution which represents spatial relationships between character components as well as randomness of handwriting. We develop a synthesis algorithm producing handwritings that exhibit the naturalness based on the proposed probabilistic model. The proposed system is applied to the Korean character synthesis. Experimental results demonstrated a high degree of visual plausibility of the synthesized handwritings. Human subjects were asked to identify the machine-generated handwritings in test images that contained both human-generated and machine-generated handwritings. The result showed that the subjects judged 89.6% of machine-generated handwritings to being human-generated in total. Contributions of this papers are 1) representation of coarticulation effect by a probablistic model, 2) synthesis of handwriting under this coarticulation model, and 3) robust handling of indiviudal handwriting style.