deep Learning for Arabic Handwriting Recognition System

Abstract

Automated handwriting recognition is a crucial element in numerous applications across diverse fields. This problem has been the subject of significant investigation during the past three decades because to its complex nature. Research has mostly concentrated on the identification and interpretation of handwritten text in Latin languages. There is a scarcity of research conducted on the Arabic language. Therefore, it is crucial to build Children's Arabic handwriting recognition software and apps. In this study, we offer three models for the recognition of children's Arabic handwriting using deep learning. In this paper, an ensemble learning is employed for the Recognition Arabic handwriting, the proposed ensemble learning combined three model, the first with three convolution layer and the second with four convolution layer and the third model is CNN using BI LSTM. Hijja, a recent collection of handwritten Arabic by children that was gathered in Saudi Arabia, is used in training process. The most relevant work achieves less accuracy from our models for the same data sets. The three models achieve accuracies 87,88, and 89, when they work independently. The performance enhanced by use the ensemble and soft voting that increases the accuracy up to 92% which better than some works selected from the literature.