Offline handwritten Uighur character recognition based on grapheme analysis

Handwritten Uighur characters contain a lot of small and random writing strokes, which make character recognition more complicated. For 128 Uighur characters, an efficient offline handwriting recognition algorithm based on grapheme (part of a character) analysis is proposed in this paper. Firstly, by dot stroke detection and component analysis, 128 character models are established by decomposing the Uighur characters as three type graphemes: dot, affix and main graphemes. Secondly, the Uighur characters are pre-classified into 12 subclasses through their grapheme compositions. Finally, different classifiers are designed for various types of graphemes. With the fusion coefficients of graphemes estimated, the character recognition result is given by fusing the graphemes classification outputs using the weighted naive Bayesian algorithm. Experimental results show that the algorithm can effectively identify the 128 unconstrained handwritten Uyghur characters.