计算机工程与应用 ›› 2014, Vol. 50 ›› Issue (11): 202-205.

• 图形图像处理 • 上一篇    下一篇

基于HMM和GMM的维吾尔语联机手写体识别研究

许  辉1,热依曼·吐尔逊1,2,吾守尔·斯拉木2   

  1. 1.新疆大学 多语种信息技术重点实验室,乌鲁木齐 830046
    2.新疆大学 信息科学与工程学院,乌鲁木齐 830046
  • 出版日期:2014-06-01 发布日期:2015-04-08

Online-handwriting recognition research of Uyghur word using GMM and HMM

XU Hui1, Riyiman TURSUN1,2, Wushour SILAMU2   

  1. 1.Key Laboratory of Multi-Language Information Technology, Xinjiang University, Urumqi 830046, China
    2.College of Information Science & Engineering, Xinjiang University, Urumqi 830046, China
  • Online:2014-06-01 Published:2015-04-08

摘要: 给出了一个基于HMM和GMM双引擎识别模型的维吾尔语联机手写体整词识别系统。在GMM部分,系统提取了8-方向特征,生成8-方向特征样式图像、定位空间采样点以及提取模糊的方向特征。在对模型精细化迭代训练之后,得到GMM模型文件。HMM部分,系统采用了笔段特征的方法来获取笔段分段点特征序列,在对模型进行精细化迭代训练后,得到HMM模型文件。将GMM模型文件和HMM模型文件分别打包封装再进行联合封装成字典。在第一期的实验中,系统的识别率达到97%,第二期的实验中,系统的识别率高达99%。

关键词: 维吾尔文, 手写体, 隐Markov模型(HMM), 高斯混合模型(GMM)

Abstract: This paper presents an online-handwriting recognition system of Uyghur language based on GMM and HMM twin-engine recognition model. In the GMM part, the system extracts 8-directional features, then generates 8-directional pattern images, locates spatial sampling points and extracts the blurred directional features. The GMM model files are formed after the iterative training of the model refinement. In the HMM part, the system obtains the line_segmen_features sequence by applying line_segment_features method. The HMM model files are got from the iterative training of the model refinement as well. The GMM and HMM model files are packaged and encapsulated respectively, and then joint-packaged into a dictionary. In the first phase of experiment, the recognition rate is 97%; in the second phase, the recognition rate increases to 99%.

Key words: Uyghur, handwriting, Hidden Markov Models(HMM), Gaussian Mixture Model(GMM)