Cursive handwriting recognition using hidden Markov models and alexicon-driven level building algorithm

摘要

The authors describe a method for the recognition of cursivelyhandwritten words using hidden Markov models (HMMs). The modellingmethodology used has previously been successfully applied to therecognition of both degraded machine-printed text and hand-printednumerals. A novel lexicon-driven level building (LDLB) algorithm isproposed, which incorporates a lexicon directly within the searchprocedure and maintains a list of plausible match sequences at eachstage of the search, rather than decoding using only the most likelystate sequence. A word recognition rate of 93.4% is achieved using a 713word lexicon, compared to just 49.8% when the same lexicon is used topost-process the results produced by a standard level buildingalgorithm. Various procedures are described for the normalisation ofcursive script. Results are presented on a single-author database ofscanned text. It is shown how very high reliability, up to near perfectrecognition, can be achieved by using a threshold to reject those wordhypotheses to which the system assigns a low confidence. At 19%rejection, 99.2% of accepted words appeared in the top two choicesproduced by the system, and 100% of the 1645 accepted words werecorrectly recognised within the top eight choices