基于组合特征的小字符集脱机手写汉字识别方法计算机系统结构专业论文.docxVIP

I I 摘 要 摘 要 脱机手写汉字有书写不清楚和不规范等特点，这给脱机手写汉字识别带来一定的 困难。本文设计了一种基于组合特征和隐马尔科夫模型(Hidden Markov models，HMM) 分类器的脱机手写汉字识别方法。 在特征选择与提取方面，本文采用了基于骨架方向特征和特征点的组合特征，首 先对单个脱机手写体汉字图像进行细化，并对汉字图像进行弹性网格划分；然后在网 格内对汉字骨架进行扫描，提取汉字的四种特征点和汉字图像的横、竖、撇、捺四种 骨架方向线素特征；最后，将汉字图像的特征点分别在整体上、横向上和纵向上和汉 字图像的四种骨架方向线素特征结合，形成整体特征、横向特征和纵向特征构成特征 向量。 在分类器设计方面，采用 HMM 分类器，利用训练集手写汉字对分类器进行训练， 采用传统投票法的平均值规则对各个 HMM 分类器的识别结果进行集成，将平均相似 值最大的类别作为识别结果。对特定的小字符集脱机手写汉字图像的识别实验验证了 该方法有效性。 关键词 脱机手写汉字识别 弹性网格 骨架方向 特征点 隐马尔科夫模型 II II Abstract Abstract It is difficult for offline handwritten character recognition because of the characteristics of unclear writing and unnormative strokes and so on. This dissertation introduces a kind recognition method of offline handwritten character based on integrated features and HMM. On the aspect of feature extraction, this dissertation adopts the integrated features based on skeleton directional characteristics and feature points. The method uses elastic mesh strategy to divide the thinned character image written by different man into meshes, then extracts four kinds of key point features and four kinds of skeleton direction features of Chinese characters image within the grid. Finally, the three feature vectors of handwritten form Chinese characters images are obtained: the overall characteristics, horizontal features and the vertical features. On the aspect of the design in the classifier, this dissertation uses HMM to classify. Based on independent HMM classifiers, we use the training handwritten Chinese set to train the classifiers. This dissertation uses voting rule to integrate the HMM classifiers and selects the maximum average similarity as the recognition results in recognition. In this dissertation, experimental results based on specified small character set show that the method is feasible. Keywords Offline handwritten character recongnition Elastic meshes Skeleton directional decomposition Feature point HMM PAGE PAGE IV 目 录 目 录 第 1 章 引 言 1 1.1 研究背景及意义 1 1.2 国内外研究现状 2 1.3 本文工作