PURPOSE: To improve a recognition ratio and recognition speed with respect to the characters of various faces and sizes by giving a uniform and stable reference position to the assembly of the feature quantities of an extracted character contour part, and extracting the topological feature quantity of the characters based on feature extraction by means of contour tracking.
CONSTITUTION: An inputted picture is raster-scanned so as to detect a contour tracking start point, and contour tracking is executed from the detected contour tracking start point so as to detect the feature quantity of the character contour part, whereby projections of the extracted feature quantity in horizontal or vertical direction at every directional code are generated. The lines for dividing the upper and lower areas or right and left areas of the characters are obtained based on the generated projections and the reference position of the character contour part feature is decided from the area division lines, whereby the feature of the character contour part is extracted. Consequently, topological information on the characters can be extracted by checking the outlines on the area division lines. Thus, said information can be used for large sort before detailed recognition, and the recognition ratio and the recognition speed can be improved.
KOJIMA KEIJI
SATO HAJIME