Recognition of Tifinagh characters using Extreme Learning Machine

The main objective of this paper is to propose a robust architecture that recognizes Amazigh language manuscript characters written in Tifinagh. To do so, we proposed first an Extreme Learning Machine (ELM) model carried on our Berber-MNIST dataset. Lack of data was a big issue in the case of our model. Hence, extending the existing database using data augmentation methods was necessary. For that, we have adopted three techniques such as duplication, brightening, and zooming. The simulation results showed that the new ELM model achieves a satisfactory results in terms of recognition rate reaching 91.42%, thanks to the use of the data augmentation techniques.