FEATURE EXTRACTING METHOD OF HANDWRITTEN CHARACTER RECOGNITION

摘要

PURPOSE:To extract features correctly, by extracting the contour line of a character from its black-and-white binary input character pattern, detecting corner points and inflection points from contour points and extracting a reference pattern, and constituting a rough partial pattern. CONSTITUTION:The black-and-white binary pattern of the input character 3 is inputted to a character pattern storage part 22 through a photoelectric conversion part 21. A contour line extraction part 23 extracts the contour line of the character and a reference pattern extraction part 24 extracts the corner points and inflection points, which are sent to a linear pattern arithmetic part 28, curved part pattern arithmetic part 29, and circular part pattern arithmetic part 30. The coordinates of a starting and an ending point, direction of a tangent at the starting point, direction of a tangent at the ending point, extent of a change in direction at a curved part, and center point and length of a circle are sent to a reference pattern storage part 25, and a reference parallel pattern part extraction part 26 finds a part parallel to the reference pattern to send the found part to a partial pattern storage part 27. The output of the partial pattern storage part 27 is sent to a recognition part.