Reconstructing Handwriting Character Font Models with Incorrect Stroke Order

Introduction

Today, displaying documents arises in the wide variety of electronic device, such as personal computer, smart phone, etc. Then various types of characters have been developed and used in such devices. The methods of designing characters are usually classified as (i) dot matrix, (ii) outline vector, and (iii) skeleton vector method, where characters are treated as planer patterns (Sakamoto, 1985; Tokura, 1988; Uehara, 1990).

On the other hand, mimicking the writing process by humans, the so-called “dynamic font method” has been developed by utilizing B-splines (Takayama, 1996). This idea is to introduce a virtual writing device and a virtual writing plane, and then characters are formed as the result of 3-dimensional motion of the writing device. Unlike the conventional methods – such as dot matrix, outline vector and skeleton vector methods, etc., this dynamic font method leads a lot of advantages in the construction and reconstruction of characters (or words and sentences). In particular, one of the advantages is to enable us to readily reconstruct characters to cursive characters as seen in Japanese calligraphy (Fujioka, 2006; Fujioka, 2008). The reconstructing scheme has been developed by employing a theory of smoothing splines (de Boor, 1978; Wahba 1990; Kano, 2005). However, such a character reconstruction has been restricted ourselves to the case where the stroke order are correct. Therefore, in such a case, we could not reconstruct the natural cursive characters from characters with incorrect stroke order.

On the other hand, some researcheres in the field of on-line character recognization have considered some problems on characters with incorrect stroke order. For example, Wakahara et al. (Wakahara, 1983) have considered a problem of recognizing some character with incorrect stroke order and incorrct stroke number. For solving such a problem, they have developed the so-called “starting point fixation method” using DP matching. However, this method is only for pattern recognition of characters, thus does not intend to modify the incorrect stroke order.

In this study, we consider the problem for reconstructing handwriting characters, where we suppose that a stroke order of characters may be incorrect. Using the modified dynamic font method, the handwriting character is modeled by employing a theory of smoothing splines. In particular, supposing that character is generated by human arm motion (Edelman, 1987; Wada, 1985; Morasso, 1982; Plamondon, 1998; Latash, 2003), we introduce a smoothing spline theory based on minimum hand jerk model (Flash, 1985). The handwriting motion Here we develop a method for correctly modifying the stroke order of characters. Such a method is developed by utilizing the modified dynamic font method and Wakahara’s “starting point fixation method”. That is, this modification process is developed as an operation on the so-called “control polygon” which is a formal representation of characters. Moreover, the computation on modification does not depend on the size of character size unlike Wakahara’s method. Using the developed schrme, it can be shown that we can reconstruct the character fonts well to natural cursive characters even when the stroke order of characters is incorrect. We examined the usefulness and effectiveness of proposed scheme by experimental studies.