Off-line hand-printed chinese character recognition based on stroke matching

The specific purpose of this thesis is the automated recognition of the off-line Chinese hand-printed characters by using a blue ball-point pen. Through mask processing, the main components in a Chinese character such as vertical, horizontal, and slant strokes can be extracted. Then, the connected components with the coordinates of the top, bottom, leftmost, and rightmost ends of each stroke extracted are found. From these coordinates, the length and position of each stroke can be computed. According to the number, relative length, and relative position of each stroke, both of the coarse and fine rule-based classification can be made, and the goal of this thesis is able to be reached. Excluding the load and segmentation of the original image, the computing time for the feature extraction and classification depends on the image size and the number of strokes. It is about 0.3 seconds per Chinese character on an IBM PC 80486 DX33. The advantages of the proposed method include efficient time complexity, strong ability to detect very similar Chinese characters, tolerance of the slope of the stroke, and 96% or higher recognition rate. The disadvantage is the inflexibility for learning driven by the users since the matching rules are open to the manufactures only at present