An area to be recognized on a document is divided into subareas in a direction of lines of characters, and the dot patterns of the characters in each of the subareas are projected substantially in the direction of the lines of the characters to extract line blocks. Those of the extracted lines which fall within a predetermined width are selected and those line blocks which overlap each other in the direction of the lines of the characters are interconnected to provide the lines of the characters from which the dot patterns of the characters are to be extracted.