基于轮廓特征的HMM手写数字识别  

Handwritten digits recognition using HMM based on contour feature

作　　者：肖明[1] 贾振红[1] 

机构地区：[1]新疆大学信息科学与工程学院,乌鲁木齐830046

出　　处：《计算机工程与应用》2010年第33期172-174,211,共4页Computer Engineering and Applications

基　　金：国家教育部新世纪人才支持计划No.NCET-05-0897~~

摘　　要：在手写数字识别中,边界链码和环构成了对字符轮廓的完整描述。针对手写数字的特点,建了24种笔划。首先将样本边界链码转化成由24个笔划组成的特征值,再加上环特征,构成整个特征值。然后利用隐马尔可夫模型(HMM)对提取的特征值进行分类识别。首次将字符轮廓特征应用在基于HMM的手写数字识别中,在识别MNIST字库上,取得了92.2%的识别率。Boundary chain code and the ring constitute a complete description of character contours in the handwritten digits recognition.24 kinds of strokes are constructed based on the characteristic of handwritten digits.Firstly the boundary chain code is transformed to a component feature by 24 strokes and together with the ring feature,which constitute the entire feature.Then the feature is recognized by using Hidden Markov Model（HMM）.The contour feature which is applied in the handwritten digits recognition of HMM is introduced firstly.The recognition rate of 92.2% is obtained from the recognition of MNIST database.

关 键 词：隐马尔可夫模型(HMM) 数字识别 边界链码 笔划 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]