Segmentation of on-line handwritten Chinese characters based on stroke merging and dynamic programming
Changsong Liu · Journal of Tsinghua University(Science and Technology) · 2004
The recognition of on-line handwritten Chinese characters has developed from a single character process to a multi-character process, and the system performance depends greatly on the segmentation result. The system performance is improved by an algorithm for on-line handwritten Chinese character segmentation. The algorithm uses geometric information to merge the strokes into character blocks based on handwriting rules and applies dynamic programming to find the best segmentation path. The width-height ratio, size, inter-character distance, and recognition information are used to construct the cost matrix. Handwriting variations are handled by a plastic algorithm that reduces the bounding box around thecharacter to exclude some parts of the character. Experiments show that the algorithm is very effective for on-line handwritten Chinese character segmentation with either numerous characters on multiple-lines or characters touchingeach other.