| 摘要 |
A method is provided for extracting bending points from character images for use in an optical character recognition procedure (28) that recognises the characters. In a preferred aspect a contour (boundary of strokes) of a character image is traced and strong curvatures are detected as bending points using heuristically determined (72) parameters with some attributes such as position, angle of curvature, convex or concave, and acuteness being provided as a data set output. <IMAGE> |
