Tunisian city name recognition based on dynamic Bayesian networks: Factorial hidden Markov model case

In this paper, we present a new approach for the recognition of Tunisian city names. This approach use two sliding windows with a uniform size in two directions (vertical and horizontal) in order to extract the features according to the lines and columns. Then, the two sequences of obtained information have been modeled by a factorial hidden Markov model. This model is a variant of the dynamic Bayesian network, in which we represent the interaction and the dependence between the two primitive sequences using an intermediate hidden layer. The approach has been tested with the benchmark IFN/ENIT database and the recorded results have been encouraging.