JPS57211675A - Character reading method - Google Patents

Abstract

PURPOSE:To reduce erroneously read characters without considerable increment of reject characters, by using only a small number of recognition candidate characters, which are obtained in the preceding recognition processing, as recognition objective characters in the second and following recognition processings. CONSTITUTION:Characters recorded on a form 1 are scanned by a scanning part 2 and are converted to a video signal, and character patterns having multilevel density are stored in a line buffer 3. A precprocessing part 4 cuts multilevel character patterns in the buffer 3 to every character, and the binary character pattern is sent to a recognizing part 5, and a recognition dictionary 6 is referred to select plural recognition candidate characters from recognition objective characters stored in a recognition character table 7. In the second and following recognition processings, plural recognition candidate characters are used as recognition objective characters. Thus, erroneously read characters are reduced without considerable increment of reject characters even in the unanimity method where a character is recognized only if all recognition results are equal to one another and the character is rejected otherwise.