Convolutional Neural Network Adoption for Offline Arabic Handwriting Scripts Identification

Abstract

      Nowadays most organizations and governments have a huge amount of handwritten documents created via their daily transactions. It became necessary to utilize computer technologies to recognize and read handwritten texts. The domain of Image processing is capable of deciding difficult problems and simplifying human actions by converting handwritten documents into digital formulas. Recognition of Handwritten Arabic alphabets has been broadly studied in previous years. This paper presents a dataset containing Arabic handwriting that can be used to evaluate the performance of an offline writer identification system. Handwritten alphabets are complex to classify because of varied human handwriting techniques, the difference in shape and size of letters, and the angle of writing. A variety of recognition methodologies for handwritten Arabic are conferred here alongside their performance. In this paper, we present-day inclusive analysis for writer identification approaches and provide a taxonomy of database, and feature extraction techniques, as a conventional classification for writer identification. This scheme attained the greatest recognition accuracy of 99.36% based on some feature extraction techniques and a Convolutional Neural Network (CNN) classifier.