Handwritten Digits Recognition Using HMM Based On Contour Feature

Boundary chain code and the ring constitute a complete description of character contours in the handwritten digits recognition. This paper extracted the boundary chain code of handwritten digits firstly, and according to the characteristics of handwritten digits, we construct 24 strokes, which are numbered from 1 to 24, every three boundary chain code will be transformed into a stroke. The ring features of digital will be ordered by no-ring, a ring and two rings sequentially numbered from 25 to 27. Those 27 numbers consisted the features of the entire sample, and then using HMM to recognize characteristics of different length parameters for classification.The contour feature which is applied in the handwritten digits recognition of HMM is introduced firstly in this paper. The recognition rate of 92.2% is obtained from the recognition of MNIST database. In the past pattern recognition, because of the existing model with the same length feature only applies to the value of recognition, of different length parameters for pattern recognition is not well studied. As the Hidden markov model can be applied to the characteristics of different length values. This paper is based on handwritten digits recognition, on the one hand, it identifies the viability in different length parameters of Markov model .On the other hand, it grasps the suitable features and holds the overall features of the character recognition, a good effect has been achieved .The last but not the least, it is also a good supplements for the only way to use local features methods.