一种基于字符HMM模型级联的手写体西文单词识别方法

A METHOD FOR HANDWRITTEN WORD RECOGNITION BASED ON CHARACTER HMM CASCADE CONNECTION

提出了一种识别西文单词的级联HMM方法,在字符HMM模型基础上按照统计语法将各模型依概率连接.它扩展了HMM的模式描述方式,允许在级联模型上表征状态的跳跃、转移和驻留等.通过共享字符模型来描述级联状态转移概率,可以更加可靠地刻画手写体单词的行为特点.采用面向级联的Viterbi算法,在完整的单词采样序列输入后直接识别,无需做字符的分割和标注,从而避免了在字典中为每个单词建立模型而导致的识别不同步问题.用EW-1单词样本库进行试验,级联模型法的第1候选识别率为89.26％,带有连字模型的HMM法的第1候选识别率为82.34％,降低错误识别率达39.18％.

In this paper, a cascade connection hidden Markov model (CCHMM) method is proposed. This model allows state transition, skip and duration, and extends the way of HMM pattern description of handwritten English words. According to the statistic syntax, it may depict the behavior of handwriting curve more reliably, while character segmenting and labeling are unneeded. The Viterbi algorithm for the cascade connection model may be applied after the whole sample series of a word is input. Compared to the method of creating models for each word in lexicon, this method could avoid the problem of recognition asynchronous. Experiments on EW-1 database shows that CCHMM approach could obtain 89. 26% recognition rate for the first candidate. The proposed approach cuts 39. 18% error rate of ligature model method, whose first candidate is 82. 34%.