A novel approach in off-line handwritten Chinese character stroke segmentation

In recognition of hand-written characters, stroke segmentation often serves as a crucial step. In this paper, we introduce a new method called manifold extraction to solve this problem. The basic idea of manifold extraction is: first build a neighborhood graph to capture the intrinsic topological structure of the sampled characters, then analyze the dimensional uniformity of neighboring points to discover the segments of strokes, finally combine the segments that are possibly from the same stroke and get the more informative structures of the characters. In this way, manifold extraction identifies the interlacing strokes in a complicated background and accomplishes the step of stroke segmentation. The experimental results show the effectiveness of this method in stroke segmentation as well as in exploratory data analysis.