Method for automatic character recognition

Abstract

Abstract From the individual characters calculated with the aid of a character classifier, character strings marked by beginning of word and end of word are formed which are checked with respect to identity or similarity with stored reference strings of a context lexicon. The context lexicon is continuously updated by continuously reading in strings containing no rejection characters at all, the repeated reading-in of identical strings being counted. Current strings are compared with the strings of the context lexicon and the best possible string with respect to similarity and frequency is selected for further evaluation. A correction provided due to the context comparison is carried out only if the substitution transition is probable due to the classifier characteristic for the characters considered in each case.

Inventors

• 👤 Helmut Dipl.-Ing. Mittelbach
• 👤 Xaver Dr.-Ing. Müller
• 👤 Peter Dipl.-Ing. Schlegel

Bibliographic Details