CASIA OpenIR  > 毕业生  > 博士学位论文
基于人工神经网络的手写汉字识别系统
王春恒
1999-07-01
学位类型工学博士
中文摘要脱机手写体汉字识别是一个典型的大类别数模式识别问题。汉字字符集具 有数量大、结构复杂、相似字多和无规则变形严重等特点,使得脱机手写体汉 字识别成为字符识别领域最大的难题和最终的目标之一。 在这种大类别识别的研究中,大多采用匹配的方法来解决问题,原因是方 法的实现比较简单。但是,基于匹配的方法分类精度较低。近年来,人工神经 网络作为一种认知模型在模式识别中得到了成功的应用,理论上已经证明网络 能够模拟Bayesian最优决策,其分布式并行信息处理机制更加符合实际情况。 另外,更重要的一点是网络的学习机制为综合集成中倡导的人机集成的实现提 供了有利的工具,这在数字识别中得到了很好的体现。 然而,在诸如汉字识别这种大类别识别问题中,利用网络实现分类和集成 并不多见。前期出现的相关报导大多采取问题分解的策略,没能完整的实现综 合集成的思想,系统的性能也没有明显的提高。 本文工作的创新之处在于完善了手写汉字识别的网络分类与集成的方法。 与以前的方法不同,原问题被作为一个整体和系统看待,每个网络都能具有独 立处理整个问题的能力。这样,就可以象解决数字识别一样来用网络解决汉字 识别问题。 值得指出的是,整个工作出发点是综合集成理论强调的人机集成的策略。 人机集成由网络的监督学习实现,能够大幅度降低原问题的复杂程度。在一个 集成系统中,针对不同层次设计不同的监督学习方式,将人的作用体现于方法 设计和系统实现不同阶段。 本文中网络分类与集成方法的提出与实现,从一定程度上完善这一方向的 研究,为人工神经网络在大类别数模式识别问题中的应用提供了一种可行的方 案。应该承认,这一工作也只是探索性的一小步而已。
英文摘要Handwritten Chinese character recognition (HCCR) is a typical large vocabulary pattern recognition problem. Handwritten Chinese characters are characterized with large vocabulary, complex structure, lots of similar characters and serious irregular variations of shapes. So HCCR is a very difficult problem and commonly regarded as one of the ultimate goals of character recognition research. In the research of large vocabulary recognition, matching approaches are frequently adopted because it is easy to be realized. But the classification performance of mateh'mg approaches is not good. In recent years artificial neural network (ANN) has been successfully applied into pattern recognition. It has been proved theoretically that ANN can simulate the Bayesian optimal decision. The distributed and parallel information processing way of ANN accords with reality. Moreover, the learning mechanism of ANN make it is very convenient to realize human-machine integration emphasized in Met synthesis. However, in case of large vocabulary recognition such as HCCR, successful work on classification and integration with AN'N is scarce. Almost all of the previous reports on this topic adopted strategy of problem decomposition, and the performance of system were not improved evidently. In this dissertation creative work is that the approach of classification and integration based on ANN is proposed. Different from previous report, the original problem is considered in a holistic and systematic way, each ANN has the capability to process the whole categories. So the problem of HCCR can be solved in the same way as numeral recognition. It is worth to be pointed out that our work starts from the strategy of human-machine integration, which is emphasized in met synthesis and realized through supervised learning. Human- machine integration can decrease the complexity of original problem. In a integrated system, different supervised learning ways are designed for different layers, and the role of man appears in different stages of method design and system realization. The presentation and implementation of the approach of classification and integration with ANN in HCCR not only improve the Chinese character recognition research, but also provide a feasible way applying ANN to large vocabulary classification problems.
关键词手写体汉字识别 综合集成 人机集成 人工神经网络 监督学习 Handwritten Chinese Character Recognition Metasynthesis Human-machine Integration Artificial Neural Network Supervised Learning
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/5703
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
王春恒. 基于人工神经网络的手写汉字识别系统[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,1999.
条目包含的文件
条目无相关文件。
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[王春恒]的文章
百度学术
百度学术中相似的文章
[王春恒]的文章
必应学术
必应学术中相似的文章
[王春恒]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。