Previous 1 [2] [3] [4] [5] [6] [7 ] [8 ]

Journal of Inforamtion Science and Engineering, Vol.8 No.4, pp.512-524 (December 1992)
Extracting Strokes of a Chinese Character Based on
Gross-Section Sequence Graph*

Jun S. Huang, Tzung Rern Jang and Fu Chang
Institute of Information Science
Academia Sinica
Taipei, Taiwan, Republic of China

Strokes are the key features of a hand-written Chinese character since a character is written stroke by stroke. In this paper, we describe a new method of extracting strokes based on a cross-section sequence graph (CSSG) proposed by Suzuki and Mori [1]. Two significant modifications on computing CSSG are proposed here. Firstly, we describe a new method of computing cross-sections by using quadratic fit of boundary points. Secondly, we use the mode of the histogram of the length of a cross section to estimate the width of a stroke and to eliminate spurious cross-sections where the width is an important parameter in constructing a CSSG. These cross-sections are then integrated into cross-section sequences (CSS's), which correspond to simple lines or curves. The remaining regions are classified as unknown regions that correspond to the end, concatenation, corner, branch, cross, and lump of strokes. These regions together with CSS's form a CSSG. Skeletons and strokes of handwritten Chinese characters are extracted from CSSG's. The experimental results show that the quality of the skeleton extracted by a CSSG is better than that obtained by using the thinning method, and that the extraction of strokes is usceesful in most cases except a few cases where the strokes are too short or distorted.

Keywords: pattern recognition, handwritten chinese characters, stroke extraction, cross-section sequence graph, skeletons

Received August 12, 1992; revised March 9, 1993.
Communicated by Wen-Hsiang Tsai.
*The previous version of this paper has been presented at the 1992 IPPR Conference on Computer Vision, Graphics, and Image processing at Taiwan, R.O.C.