Unconstrained handwritten Chinese string recognition system for the amount on bank checks

The large variations among the handwritten Chinese characters for the amount on bank checks make segmentation and recognition very difficult. However, a new Chinese string recognition method was developed based on the Hidden Markov Model (HMM) which incorporates both segmentation and recognition. In the segmentation, the Chinese character segment was the basic segmentation unit with consideration of the overlap and conglutination in handwritten Chinese character strings. The accuracy of the Chinese string recognition is greatly increased by using multi -selection candidates and dynamic programming to optimize the final recognition result.