|
In this thesis, a HMM based approach is presented to recognize the handw
ritten words of mcdical record. The proposed approach is based on the us
ing of generating a loose segmentation of a word image into a state sequ
ence and to match the sequence to a lexicon of candidate strings by a si
ngle hidden Markov model (HMM). A HMM is a doubly stochastic proccss wit
h an underlying stochastic process that is not observable, but can only
be observed through another set of stochastic processes that produce the
observable sequence of symbols. It has the advantages that it does not r
equire segmentation of word into characters and is quick to be trained.O
nce the model is established, the recognition algorithm, known as Viterb
i algorithm, iscmployed to recognize a best optimal state sequence in th
e lexicon. If not, we use the hypothesis generation scheme to find the h
ighest probability word in the lexicon. Experimental results reveal the
feasibility and efficiency of the proposed approach in recognizing handw
ritten words of medical records.
|