基于可训练机制的联机维吾尔手写字母识别技术研究  

ON TECHNOLOGY OF TRAINABLE MECHANISM BASED ONLINE UYGHUR HANDWRITTEN LETTER RECOGNITION

作　　者：木塔力甫.沙塔尔 李春庚[1] 艾斯卡尔.艾木都拉[2] 安居白[1] 

机构地区：[1]大连海事大学信息科学技术学院,辽宁大连116026 [2]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《计算机应用与软件》2011年第9期41-44,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(60772118)

摘　　要：在深入研究英文和汉字手写识别的基础上,结合维吾尔文字母的特点,提出一种基于支持向量机机器学习算法的维吾尔文联机手写字母识别方法,系统研究了样本采集、预处理、特征提取和分类等模块。在预处理中,为了消除干扰和噪声及比较中的相似性,采用了平滑滤波和线性归一化处理;考虑到维吾尔文相似字母较多,为了有效提取特征,将结构特征和统计特征相结合,提取了字符的梯度方向特征;分类器采用支持向量机。实验表明,随着训练样本的增加,识别率可以从90.62%提高到96.09%。In this paper,through in-depth study on the state-of-the-art of on-line Chinese and English handwriting recognition technology,and in combination with the unique shape of Uyghur letters,the authors proposed a method for online Uyghur letter recognition based on support vector machines,and researched systematically its modules including data sampling,pre-processing,feature extraction and classifier,etc.In the pre-processing,in order to eliminate the noise and the similarity in comparison,the authors use smoothing filtration and linear normalisation.Taking into account that there are more similar letters in Uyghur alphabet,the authors use the method of gradient directional feature of the character for effective feature extraction,which combines the structural features and statistical features together.The classifier uses support vector machines,a branch of statistical learning theory.Experimental results show,with the increase of the training data,the recognition rate increases from 90.62% to 96.09%.

关 键 词：支持向量机 联机手写识别 维吾尔文字母 归一化 特征提取 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]