Arabic Braille Numeral Recognition Using Convolutional Neural Networks

Abstract

Braille is a system that is designed to assist visually impaired individuals to acquire information. It consists of raised dots arranged in a cell of three rows and two columns. Visually impaired individuals rely on the sense of touch to read and write. However, it is difficult to memorize the arrangement of dots that compose a character. This research aims to design an application that recognizes and detects Arabic braille numerals and convert it to plain text and speech by implementing convolutional neural network variation Residual Network (ResNet). A new dataset was collected by capturing Arabic braille numerals using smartphone cameras. The recognition accuracy for Arabic braille numerals achieved 98%, taking into accountability different light and distance conditions.