PERFORMANCE IMPROVING METHOD OF GRAPHEME BASED HANGUL RECOGNIZER

摘要

Determine whether the present invention relates to a perilla-based character recognizer performance enhancement, the method comprising: classifying basil combination type for a character image included in the document image, on the basis of the classification basil combination type it includes a trailing in the text image granulomatous or without determination step, basil classification step of classifying the basil included in the character image on the basis of the determination result of the trailing presence determination step, on the basis of the determination result of the trailing or absence decision step of herbicidal classified in the basil classification method comprising: but recognize the consonant, a medial vowel and / or trailing consonant, a medial vowel recognition is a leading consonant, a neutral recognized by the Phoneme recognition step, the Phoneme recognition steps executed on the basis of the recognition reliability of the leading consonant and / or trailing consonant and a collection and / or basil in combination the steps of combining a recognition result of the trailing consonant. Therefore, so that the leading consonant and a trailing by using the consonant recognition information to a medial vowel recognition it is possible to increase the collection of recognition, the sum of recognition confidence value of a leading consonant and a trailing consonant first one to recognize the bar relatively recognized as difficult neutral by multiplying the maximum value of the bar, and then normalizing the neutral characteristic values ​​normalize the consonant recognition confidence values ​​and the collection featured value value is distributed in mutually different effective cycle of the recognition of the neutral bar, not biased to the characteristic value the value of the credit value or set of consonant because there is an effect that it is possible to obtain a stable recognition.