| 摘要 |
<p>Handwritten text recognition uses flexible text model taking into account occurrence probabilities of series of letters. Symbol models are used e.g. hidden Markov model. User inputs handwriting by digitizer tablet or onto display screen, pen input is detected and values and characteristics vectors stored in memory. User selects recognition mode for handwriting recognition. Two modes available, multi word recognition mode with recognition proceeding over the word boundaries to recognize entire sentence. In second mode, single word recognition, isolated recognition of single words takes into account additional recognition of punctuation marks. Multi-word recognition uses prestored text model (106) held in memory and hidden Markov model (17) trained in previous training phase. Text model describes occurrence probability of a letter under condition of specific series of letters before the letter and determines occurrence probability for a series of characters. To train text model for multi word recognition, entire sentence is used to determine statistic relationships over word boundaries. Second recognition mode, for single words, has further step of single word recognition. It uses another text model for single word recognition (109) and hidden Markov model for single word recognition (110). Independent claims included for data processing unit, computer readable storage medium for program for recognition.</p> |
