RIA: Handwritten Word Recognition Using First and Second Order Hidden Markov Model

隐马尔可夫模型（HMM）是一个双重随机过程，其底层随机过程是不可观察的，即状态是隐藏的，但只能通过另一组随机过程来观察，这些随机过程产生可观察的符号序列。本文在HMM框架下对手写体识别问题进行建模。对于本文重点研究的英语文本，用字母表中的字母来识别状态，并通过实验研究产生最优符号。这项任务使用了15个特性（有些是旧的，有些是新的）。一阶和二阶隐马尔可夫模型用于识别任务。利用现有的英语统计学知识，如密码学等，极大地简化了一阶模型参数的计算方案。扩展这些结果，并通过穷举字典搜索，计算二阶模型的概率。Viterbi算法用于识别单个最优状态序列，即组成单词的字母序列。本文还研究了对识别算法的修改以适应上下文信息。最后给出了一阶和二阶模型的实验结果。

A hidden Markov model (HMM) is a doubly stochastic process with an underlying stochastic process that is not observable, i.e., states are hidden, but can only be observed through another set of stochastic processes that produce the observable sequence of symbols. In this work, the handwritten script recognition problem is modeled in the framework of HMM. For English text, which is the focus of the present research, the states are identified with the letters of the alphabet, and the optimum symbols are generated by means of experimental study. Fifteen features (some old, some new) are used for this task. Both the first and second order hidden Markov models are used for the recognition task. Using the existing statistical knowledge of the English language as in Cryptography etc., the calculation scheme of the model parameters are immensely simplified for the first order model. Extending these results and through an exhaustive dictionary search, probabilities of the second order model are calculated. Viterbi algorithm is used to recognize the single best optimal state sequence, i.e., sequence of letters consisting the word. The modification of the recognition algorithm to accommodate context information is also researched. Finally, experimental results for the first and second order models are provided.