Handwritten similar Chinese characters recognition based on multi-class pair-wise support vector machines

The existence of lots of similar characters is a key factor affecting classification performance of off-line handwritten Chinese characters recognition. This paper studies on handwritten similar Chinese characters recognition problem based on support vector machines. The features of normalized Chinese character are extracted by the method of wavelet transform and elastic meshing. The classification performances of similar Chinese characters are compared based on max voting, fuzzy pair-wise SVM and directed acyclic graph.