摘  要
随着电子信息技术的迅速发展，人们越来越多地使用计算机等电子设备进行文字数据处理，键盘输入、触屏手写、数位板书写等电子输入方式也逐渐在部分场景中取代传统的纸笔书写。然而，今天汉字的传统纸笔书写仍然是汉语学习过程中的重要组成部分，在练习汉字书写的过程中，需要关注汉字书写的质量，做到端正、整洁、优美。但是人工方式的书写评分存在主观性强、工作成本高等缺点，无法满足庞大用户群体的需求。
本课题的目标是探究在摄像头监测的半联机状态下手写汉字提取及书写评分的有关技术，结合当前神经网络理论思想，设计和实现一个切实可行的手写汉字书写评分系统，以更好地帮助汉语学习者练习汉字书写，提高书写质量。本论文的主要工作如下：
1)针对本论文对手写汉字书写评分的需求，提出了基于形态学处理的半联机手写汉字笔画提取算法，利用视频帧序列的前后差异筛除图像中的非文字区域，同时识别出不同笔画之间的切换点，降低了笔、手部等噪声的干扰，提高了笔画识别的准确性。
2)针对本论文的工程应用需求，搭建了一个切实可行的手写汉字书写评分系统。该系统分为两大部分：手写汉字笔画提取、手写汉字书写评分。该系统可以实现的功能为：首先，从汉字书写视频中自动分别提取出汉字的所有基本笔画；其次，对每一个笔画的类型进行识别；最后，给出书写质量的综合评分。本文对系统各个模块进行了测试，展示并分析了实验结果，通过实验验证了该系统对自然场景下的手写汉字提取和书写评分的有效性。
关键词：卷积神经网络；Android系统；半联机手写汉字识别；汉字书写质量评分
Abstract
With the rapid development of electronic information technology, people are increasingly using electronic devices such as computers for text data processing. And electronic input methods such as keyboard, touch screen and digital pad are gradually replacing traditional writing method with pen and paper in some scenarios. However, today traditional writing method of Chinese characters is still an important part of the Chinese learning process. In the process of practicing writing Chinese, we should focus on the quality of Chinese characters, and try to keep in a correct, neat, and graceful manner. Yet the manual way of writing scoring has the disadvantages of being highly subjective and expensive to work with, and cannot meet the needs of a large group of users.
The goal of this project is to investigate the techniques related to handwritten Chinese character writing scoring, and to design and implement a practical handwritten Chinese character writing scoring system by combining current neural network theoretical ideas, in order to better help Chinese learner practice Chinese character writing and improve writing quality. The main work of this thesis is as follows.
1.For the needs of handwritten Chinese character writing scoring in this thesis, and online handwritten Chinese character stroke extraction algorithm based on morphological processing is proposed, which uses the before-and-after differences of video frame sequences to filter out non-text regions in the images. Then identifies the switching points between difference strokes, as well as reducing the interference of noises of pen and hand, which improves the accuracy of stroke recognition.
2.A practical handwritten Chinese character writing scoring system is built for the engineering application needs of this thesis. The system is divided into two major parts: handwritten Chinese character stroke extraction and handwritten Chinese character writing scoring. The system can achieve the following functions: firstly, automatically extract all the basic strokes of Chinese character separately from the writing video; secondly, identify the type of each stroke; finally, give a comprehensive scoring to the writing quality. In this paper, we test each module of the system, show and analyze the experimental results, and verify the effectiveness of the system for handwritten Chinese character extraction and writing scoring in natural scenes through experiments.
Keywords: CNN；Android system；semi-Online-HCCR；Chinese character writing quality score
目  录
摘  要 I
Abstract II
目  录 IV
第1章 绪论 1
1.1. 研究背景与意义 1
1.2. 国内外研究现状 2
1.2.1. 文本定位技术研究现状 2
1.2.2. 手写汉字识别研究现状 3
1.2.3. 汉字书写质量评价方法研究现状 4
1.3. 本文所做工作 4
1.4. 论文结构安排 5
第2章 系统需求分析及概要设计 6
2.1. 可行性分析 6
2.2. 系统用例分析 7
2.2.1. 用例表设计 7
2.3. 功能需求分析 9
2.4. 性能需求分析 10
2.5. 系统总体设计 10
2.5.1. 系统总体架构设计 10
2.5.2. 系统功能模块设计 12
2.6. 相关技术介绍 13
2.6.1. Android系统 13
2.6.2. OpenCV视觉库 14
2.6.3. TensorFlow开源库 14
2.7. 本章小结 14
第3章 系统详细设计及实现 16
3.1. 系统模块详细设计 16
3.2. 笔画提取模块 16
3.2.1. 单帧图像的疑似笔画像素点检测 17
3.2.2. 基于帧差法的笔画增长点检测 19
3.2.3. 笔画切换点检测与单笔画的提取 20
3.3. 基本笔画识别模块 22
3.3.1. 汉字的基本笔画 22
3.3.2. LeNet-5网络 23
3.4. 书写评分模块 25
3.4.1. 重心特征 27
3.4.2. 网格特征 27
3.4.3. 整字综合评分 27
3.5. 系统UI界面 28
3.6. 本章小结 30
第4章 系统性能测试 31
4.1. 测试的意义与方法 31
4.2. 运行性能测试 32
4.2.1. 测试设备 32
4.2.2. 测试环境 32
4.3. 本章小结 33
总结 34
1. 论文工作总结 34
2. 工作展望 34
参考文献 35
致谢 37