Handwriting Digits Recognition Using New Neuro-markovian Approach

Résumé: This paper presents a hybrid approach based on the Multi Layer Perceptron (MLP) neural network and the Hidden Markov Model (HMM), in addition of some feature extraction methods for the recognition of handwriting digits. The features extraction methods are composed of the entropy method, the object follow-up method, the dropping method, the average line, the column method and finally the descriptors method of the normal Gaussian law. Initially, the learning for each recognition system is used by a 2000 images database where the MLP neural network is based on the gradient backpropagation learning algorithm, and the HMM system is based on the Baum-Welch backpropagation learning algorithm. Furthermore, for each primitive vector, the number of the recognized digits is calculated by the MLP system in which the non-recognized digits are tested by the HMM system. Finally, the recognized digits obtained by both systems are summed for the calculation of the recognition rate. The application of the proposed approach using Matlab gave an impressive recognition rate of 98.8%

Publié dans la revue: Journal of Advanced Research in Science and Technology