Handwritten Nushu Character Recognition Based on Hidden Markov Model

This paper proposes a statistical-structural character learning algorithm based on hidden Markov model for handwritten Nushu character recognition. The stroke relationships of a Nushu character reflect its structure, which can be statistically represented by the hidden markov model. Based on the prior knowledge of character structures, we design an adaptive statisticalstructural character learning algorithm that accounts for the most important stroke relationships, which aims to improve the recognition rate by adapting selecting correct character to the current handwritten character condition. We penalize the structurally mismatched stroke relationships using the prior clique potentials and derive the likelihood clique potentials from Gaussian mixture models. Theoretic analysis proves the convergence of the proposed algorithm. The experimental results show that the proposed method successfully detected and reflected the stroke relationships that seemed intuitively important. And the overall recognition rate is 93.7 percent, which confirms the effectiveness of the proposed methods.