Two-dimensional neural networks for handwritten Chinese character recognition

A two-dimensional Hopfield network approach is proposed to solve the handwritten Chinese character matching problem. The Hopfield net can solve the problem even if the number of strokes in the unknown character and the model character are different. In the recognition stage, the matching rates between the input character and each model character in the database are computed and used to indicate which one is the best match. The proposed technique provides a more general formulation such that some difficult issues in Chinese character recognition like rotational and translation invariance problems are solved. Theory shows that the proposed scheme requires fewer heuristics than other methods. Experimental results are reported using both synthetic and real handwritten Chinese characters to corroborate the theory