Feature extraction by hierarchical overlapped elastic meshing for handwritten Chinese character recognition

Meshing is a popular technology for feature extraction in handwritten Chinese character recognition. In this paper, we propose a hierarchical overlapped elastic meshing (HOEM), which integrates overlap, fuzzy membership and hierarchical division into elastic meshing. The overlapping zones and the fuzzy memberships of the pixels in each zone are dynamically calculated through mapping the input image to a virtual normalized image. Such mapping is fulfilled by nonlinear shape normalization based on weighted dot density. Furthermore, using hierarchical division, the global overlapped elastic meshing (GOEM) is extended to the proposed HOEM. Experiments are conducted on the ETL9B database to compare different meshing methods. The results indicate that the proposed method achieves a higher recognition rate than conventional meshing methods.