Deep morphological gradient for recognition of Arabic alphabets

Abstract

The recognition of Arabic handwritten characters by deep learning algorithms has experienced a remarkable movement in the last three years. This change can enrich the field of optical character recognition (OCR). We present in this project a deep morphological gradient for the handwritten characters recognition problem of the Arabic language based on a multilayer perceptron architecture (MLP) preceded by the morphological gradient algorithm to detect the outlines of the alphabets. This model is applied to the database of Arabic manuscript characters available on Kaggle which consists of 16,800 images. The classification accuracy of the model was 99.9% with a very minimum loss of 0.3%.