Curved segmentation path generation for unconstrained handwritten Chinese text lines

A new method of generating curved segmentation paths of handwritten Chinese text line is proposed. For overlapped characters, instead of background thinning, analysis of connected components in the overlap region of two neighboring characters is carried out, and a curved segmentation path is generated by connecting a series of points that separate the connected components belonging to different characters. As for touched characters, the candidate points are identified by using the histogram and skeleton information. After they are split at all the candidate points, curved segmentation paths are generated the same way as in overlapped characters. The experimental results on the unconstrained handwritten offline Chinese text lines show that our algorithm can improve the correct segmentation rate by 3.7% and valid segmentation rate 10.7%, indicating its effectiveness.