ACCELERATING HANDWRITTEN BERBER LATIN SCRIPT RECOGNITION USING FAST EXTREME LEARNING MACHINE

Abstract

Handwritten Character Recognition (HCR) has garnered significant attention due to its wide range of applications. This paper focuses on the application of Extreme Learning Machines (ELM) for recognizing Berber Latin manuscript characters. The Berber language is an Afro-Asian language spoken across North Africa, and HCR systems are needed to promote the digitization of old Berber documents for data sharing and heritage preservation. However, research on Berber character recognition is limited, mainly due to the existence of various writing styles and its unified writing system (Tifinagh, Latin, and Arabic).

In this work, an HCR system based on the ELM technique is developed, utilizing the Berber-MNIST dataset created in previous work. The training phase's execution time is reduced compared to previous approaches, and feature selection is performed using variance thresholding to enhance classification accuracy. Simulation results demonstrate improved classification accuracy and reduced errors. This is the first time the recognition of the Amazigh handwritten alphabet is addressed using the ELM approach.

The paper proceeds with a review of related works on handwritten character recognition based on the ELM paradigm. An enhanced ELM algorithm is discussed, followed by the presentation of the proposed architecture. A comprehensive evaluation of the ELM algorithm on Berber-MNIST datasets is provided. Finally, the paper concludes with a summary of the findings and future perspectives.