In this paper

a new approach for handwritten Chinese character recognition based on multi-channel PCA (principal component analysis)model is proposed. In terms of the stroke directional characteristics of the handwritten characters

a handwritten Chinese character is decomposed into the four directional sub-patterns at first

namely

horizontal (一)

vertical(丨)

left up diagonal (丿) and right up diagonal( )sub-pattern

each of which could be modeled by its principal components. Then

based on their four sub-pattern PCA models

a multi-channel PCA model for each category of the handwritten Chinese character is constructed respectively

and the model's reconstruction error is used as a matching measure for the handwritten Chinese character recognition. The method can not only exploit principal components' ability for representing the handwritten Chinese character sample set

but also effectively reduce the training time for modeling. Experimental results on 1034 categories of handwritten Chinese characters indicate that

the proposed method can improve recognition rate by 4.4% comparing to the Euclidean distance classifier

while its training time is much lower than that for modeling handwritten Chinese character directly by its PCA model

showing the effectiveness of the proposed approach.