To search, Click below search items.


All Published Papers Search Service


Neural Network Classifiers for Off-line Optical Handwritten Amazighe Character Recognition


Mohamed Abaynarh, Hakim Elfadili, Khalid Zenkouar, Lahbib Zenkouar


Vol. 12  No. 6  pp. 28-36


Recognizing Amazighe characters is a difficult task in the area of optical character recognition (OCR). This paper describes a new hybrid Amazighe character recognition system based on an artificial neural network classifier using Legendre moments without any preprocessing. The features extraction stage uses a set of moment descriptors which are invariants under shift and scaling. The actual classification is done using a multilayer perceptron network with learning algorithm to generate a near optimal feed forward neural networks dynamically for the task of object recognition. The proposed crossover operators aim to adapt the networks architectures and weights during the evolution process. To evaluate our proposed model a real-world database of Amazighe handwritten characters containing 7524 handwritten character images is used. This new database has been developed at the Communication an Electronics Laboratory of EMI (Ecole Mohamadia d’Ingenieurs). Experiments using our database demonstrate that combining features moments with neural network classifiers indeed are far more effective. Evaluating the proposed system with 924 test samples the recognition rate of 97.62% is achieved.


Character recognition, Amazighe characters, Legendre moments, neural networks