An Intelligent System for Chinese Calligraphy

Our work links Chinese calligraphy to computer science through an integrated intelligence approach. We first extract strokes of existent calligraphy using a semi-automatic, twophase mechanism: the first phase tries to do the best possible extraction using a combination of algorithmic techniques; the second phase presents an intelligent user interface to allow the user to provide input to the extraction process for the difficult cases such as those in highly random, cursive, or distorted styles. Having derived a parametric representation of calligraphy, we employ a supervised learning based method to explore the space of visually pleasing calligraphy. A numeric grading method for judging the beauty of calligraphy is then applied to the space. We integrate such a grading unit into an existent constraint-based reasoning system for calligraphy generation, which results in a significant enhancement in terms of visual quality in the automatically generated calligraphic characters. Finally, we construct an intelligent calligraphy tutoring system making use of the above. This work represents our first step towards understanding the human process of appreciating beauty through modeling the process with an integration of available AI techniques. More results and supplementary materials are provided at http://www.cs.hku.hk/∼songhua/calligraphy. Introduction and Related Work The field of computer vision is predominantly concerned with recognizing shapes and meanings of objects by their images. But there is more than just knowing what things are. In everyday life, our visual perception also leads to a sense of how beautiful things are. Recently, there is an increased interest in affective computing (Picard 2000).To the best of our knowledge, however, there has not been a solution to beauty appreciation by numerical means. Turing once said in his innovational paper (Turing 1950) that “We do not wish to penalise the machine for its inability to shine in beauty competitions, ...” But witnessing the advancement of photorealistic techniques in computer graphics nowadays, ∗Contact him at [email protected]. Copyright c © 2007, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. if computers can tell the beautiful from what is not, beauty contests could certainly be open to them as well. Such contests will probably not feature a catwalking computer, but rather computer-generated images of beauty, or these images against real, human-generated ones. This might be going overboard, but to imbue the computer with the ability to recognize beauty will likely win general support. Furthermore, the ultimate intelligent machine in people’s mind is one that can create results of beauty on its own, which certainly represents a nice challenge for AI researchers. We feel entailing the computer in understanding or even producing outputs of beauty that are visible, such as painting, calligraphy and sculpture, is a reasonable first step in the long journey to arrive at the ultimate intelligent machine that can deal with beauty of any form. This paper presents such a “first step” of ours—an approach to the problem of understanding the beauty of Chinese calligraphy and its facsimile by the computer. We picked Chinese calligraphy because of its great importance in Chinese art and history and the many interesting challenges it presents to the computer. Our solution demonstrates the power of integrated intelligence. The closest related work to this paper is the automatic artistic Chinese calligraphy generation system by Xu et al. (Xu, Lau, & Pan 2003; Xu et al. 2004; 2005). Their work, however, is concerned mainly with using constraint based reasoning to generate stylistic calligraphic characters, and paid very little attention on the aesthetic aspects of the generated results, a major issue to be addressed by our paper. Avizzano et al. (Avizzano, Solis, & Bergamasco 2002) proposed a calligraphy tutoring system using a haptic interface. In the field of computer graphics, there has been some work on automatic painting creation which however is mostly done with reference to a given photograph (Hertzmann 2003). Others have explored using a combination of artificial intelligence and interactive techniques to produce painting-style animation (Xu et al. 2006). The book by McCorduck (McCorduck 1990) provides a comprehensive treatment on artificial intelligence art. Outside the domain of visual arts, computer music is probably the single most successful research area in which AI techniques have been employed to do or assist