A Preliminary Attempt To Artistic Chinese Calligraphy Generation

This thesis takes a first attempt to bridge the aesthetics of Chinese traditional callig-raphy and the modern computer technology of artificial intelligence. This work is inspiredby our dreams of 1) training a computer for learning the aesthetic feelings from humansand being able to enjoy the fine arts itself and indite new artworks; and 2) Computerizingthe Chinese calligraphy, one of the key forms in Chinese traditional arts and importantroles in Chinese History, inheriting it and developing in our information-based world.We propose a hierarchical ellipse-based parametric model for representing the callig-raphy writing, which approximately simulates the behaviors of a moving brush on paper.In order to derive characters in this model from copybooks or CAD systems of virtualbrushes, a two-phase semi-automatic mechanism is used for decomposing strokes fromthe existent calligraphy images. For most regular styles of samples, in the first phasea best probable extraction based on heuristic search is applied to find a stroke-to-strokematching with their corresponding standard font; for the difficult cases in highly random.cursive, or distorted styles, the second phase presents an intelligent user interface to allowthe user to provide input for helping the extraction process.Having achieved hundreds of calligraphic characters in their parametric representa-tion, we employ a bunch of machine learning approaches based on probabilistic scales tostudy variety of calligraphy samples with human's grades. Features in both the shape ofindividual strokes and the spatial relation between strokes are considered, plus the styleconsistency among all strokes in a character. After the iterative training process, the com-puter is able to evaluate more calligraphy samples. Three probabilistic values are given for each new character, denoting its visually aesthetic quality.Once an automatic calligraphy visual quality grading functionality is available, wecan integrate it into an existent automatic Chinese calligraphy generation system. It im-proves the visual quality of generated calligraphy significantly. Experiment results showthat our automatic calligraphy aesthetics evaluating algorithm can yield scores fairly sim-ilar to those given by the human experts. With its help effectively, the calligraphy gener-ation system can produce good-looking calligraphy automatically. This work representsour first step towards understanding the human process of appreciating beauty throughmodelling the process with an integration of available AI techniques.