部件字典结合时分方向特征的手写维吾尔字符识别

针对联机手写维吾尔128类变体字符,提出了一种基于部件字典和时分方向特征的识别算法。该算法首先结合连笔分析,将字符分解为主体、附加和点三类部件,建立手写维吾尔字符的部件字典,有效解决联机手写维吾尔字符的笔顺连笔自由问题;然后,为减轻手写字符拓扑变形造成的干扰,对单个部件提取一种新的联机特征-时分方向特征,该特征在模糊域提取方向链码然后按时序划分统计,并通过检测和调整短时变动进行抖动校正;最后,设计不同分类器对各部件进行匹配,利用匹配测度分布估计各部件权重,并通过加权朴素贝叶斯融合得到字符识别结果。实验结果表明,该算法能有效地识别128类无约束手写维吾尔字符,在包含13 056个样本的手写体维吾尔字符数据库上的平均识别率为93.15%。

Handwritten Uyghur character recognition based on radical dictionary and time division direction feature

For 128 Uyghur characters,a handwritten recognition algorithm based on radical dictionary and time division direction feature is proposed.First,the radical dictionary is established by decomposing the Uyghur characters as three type radicals,main,affix and dot.The problem of stroke order and connection can be solved by analysis of the connected and broken strokes.Second,a new online statistical feature,named time division directional feature,is extracted from every radical to reduce the interference from handwritten topological deformation.The statistical feature is designed by dividing freeman codes according to time sequence,with its short term variations being found and adjusted to correct the dither phenomenon.Finally,different classifications are designed for various types of radicals.With the radical coefficient estimated according to the coefficient distribution,the character recognition result is obtained by fusing the outputs of all classifications using the weighted naive Bayesian algorithm.Experiment results show that the algorithm can effectively identify 128 unconstrained handwritten Uyghur characters.An average recognition rate of 93.15% is achieved on the Uyghur character database containing 13056 samples.