基于深度学习的汉字书写风格建模与重建方法研究

项目名称： 基于深度学习的汉字书写风格建模与重建方法研究

项目编号： No.61472015

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 肖建国

作者单位： 北京大学

项目金额： 80万元

中文摘要： 近年来，汉字书写危机引起国人的广泛关注与警觉，随着计算机与智能移动设备的使用和普及，人们的汉字书写能力显著下降。主要原因之一是，当前除了手写输入外其他面向汉字书写的各种应用的研究与开发还不成熟，在信息技术带来的无纸化趋势的冲击下，需要手写汉字的场合越来越少。本项目拟通过对汉字书写风格建模与重建机理的深入研究，解决包括海量手写汉字数据挖掘、汉字书写轨迹精确匹配、汉字书写风格特征提取等关键技术，提出适用于汉字书写风格建模的深度人工神经网络的构建与训练方法，结合最优形状插值等技术来实现汉字书写风格的精确重建，并使用该风格建模与重建方法来显著提升面向汉字的笔迹鉴定、数字墨水和手写体字库自动生成系统的应用效果。本项目的研究一方面有望解决采用现有方法难以处理的汉字书写风格的建模与重建难题；另一方面将推动汉字书写在计算机与移动终端上的使用与流行，具有重要的理论和应用价值，预期将产生显著的社会与经济效益。

中文关键词： 深度学习；建模；重建；汉字；书写风格

英文摘要： In the last few years, the Chinese handwriting crisis has attracted lots of attentions in China. With the popularization of personal computers and smart mobile devices, the Chinese handwriting capability of many people has declined sharply. One major reason is that, except for handwriting input techniques, the research and development of other Chinese handwriting based applications are still immature. Thus, the number of situations that need to write Chinese characters becomes less and less. By investigating the modeling and reconstructing principles of Chinese handwriting styles, this project intends to resolve several key techniques including the data mining of massive Chinese handwriting data, accurate registration of Chinese handwriting trajectories, feature extraction of Chinese handwriting styles, and so on, and to propose new construction and training methods for the deep artificial neutral network that is suitable for the modeling of Chinese handwriting styles. Moreover, this project also wants to integrate the optimal shape interpolation and other related algorithms to precisely reconstruct Chinese handwriting styles, and then applies the proposed modeling and reconstructing methods to significantly enhance the performance of writer identification, digital inks, and handwriting fonts generation systems for Chinese characters. Researches of this project on the one hand could address the unsolved problems about the modeling and reconstruction of Chinese handwriting styles, on the other hand could boost the utilization and prevalence of Chinese handwriting on computers and mobile devices. Our researches are of great importance in both theories and real applications, and are expected to have significant impacts on the society and economy.

英文关键词： Deep Learning;Modeling;Reconstruction;Chinese Characters;Handwriting Style