Offline Arabic Handwriting Recognition Using a Deep Neural Approach

Abstract

Arabic handwritten recognition systems face several challenges such as the very diverse scripting styles, the presence of pseudo-words and the position- dependent shape of a character inside a given word. These characteristics complicated the task of features extraction. The paper presents a deep neural approach for the handwritten recognition of Arabic words. This work is focusing on the offline recognition, thereby, the processed information represents an image. We chose the CNN method, which is one of the deep architectures which permits to remove several steps from the recognition process, including preprocessing and feature extraction. The used database is NOUN v3 contained images represented the Algerian cities. A CNN architecture was trained and then tested on the database to accomplish this task. The advantage of a CNN is that it can extract specific features from each image while compressing it to lower its initial size. Our experimental study, gives a satisfactory word recognition rate.