Comparison of shape-based and stroke-based methods for segmenting handwritten Chinese characters

The segmentation of handwritten Chinese text into Chinese characters is an important preprocessing step to the offline Chinese character recognition. It is also a very difficult task due to so many Chinese characters and their handwritten structures can be very complex. Many researchers have developed various algorithms during the past decade by Darning Shi, R. I. Damper, S. R. Gunn (2003), L.Y. Tseng, C.T. Chuang (1997), L.Y. Tseng, R.C. Chen (1997), Feng Lin, Xiaoou Tang (2002), Han Zhang, Chao Lu (2004), S.Y. Zhao, Z.R. Chi, P.F. Shi and Q. Wang (2001). In this paper, we compare two of the existing algorithms. The first one is spatial shape-based algorithm proposed by Han Zhang, Chao Lu (2004), which segments the character strings into radicals, not dealing with stroke identification. The second algorithm is stroke-based by L.Y. Tseng, C.T. Chuang (1997), L.Y. Tseng, R.C. Chen (1997), Feng Lin, Xiaoou Tang (2002), which traces each stroke and draws the stroke-bounding box, then merges the boxes by a set of rules. Based on the algorithms presented by L.Y. Tseng, C.T. Chuang (1997), L.Y. Tseng, R.C. Chen (1997), Feng Lin, Xiaoou Tang (2002), Han Zhang, Chao Lu (2004), we wrote C++ programs for time complexity, accuracy performance comparisons using different handwritten Chinese character texts. Our experimental result shows that the spatial shape-based algorithm by Han Zhang, Chao Lu (2004) is faster and more accurate.