Strokes Extraction Of Off-line Handwritten Chinese Characters

Writer identification is recognizing Chinese characters written on paper automatically with the help of computer, it has become a white-hot research point in pattern recognition which is largely different from Printed Chinese Character Recognition and On-line Handwritten Chinese Character Recognition. Compared with Printed Chinese Character Recognition, the style of handwritten Chinese characters is diverse and optional which is hard to find rules in it. On the other hand, compared with On-line Handwritten Chinese Character Recognition, writer identification doesn't have any real-time information. As stroke is primitive of Chinese character, stroke extraction is a significant step in writer identification. However, stroke extraction is still one of the most challenging research topics because it involved a large number of characters, considerable writing style and complex structure. According to the speciality of writer identification, new algorithms should be proposed.First, in order to recognize handwritten Chinese characters, dividing document image into individual characters is a very important step. Upon that density-based spatial clustering is proposed which treats with every pixel and looks at the cluster as a separate character. Also, simulation results prove it can deal with the problem of characters adhered to each other effectively. To sum up, the algorithm provides an efficient method to partition connective strokes.And then, thinning character images may bring not only the elimination of image information redundancy but also the predigestion of character recognize. Thinning procedure is the preprocessing step of character recognize. Therefore, an optimized thinning algorithm is proposed. Experiment results show that it can improve on dealing with the local configuration distortion and spurious segments. In addition, thinning algorithm based on tracing-contour is introduced to resolve the problem of local configuration distortion.At last, ambiguous zones are places where strokes intersect or overlap, detecting them and analyze the strokes near them is helpful to extract strokes. By analyzing the corresponding relation between the fork points on skeleton and ambiguous zones, an algorithm to detect ambiguous zones is proposed. The experiment results prove that it can detect ambiguous zones correctly, and then, sub-strokes besides the ambiguous zones are analyzed and merged.