An intelligent guide to author identification for Persian scripts in handwritten tasks

Seyed Hesamoddin Hashemi¹ Alireza Naghsh² Fatemehalsadat Beheshtinejad³ Mojarrad Ghahfarokhi⁴

1) MSc Student in Artificial Intelligence and Robotics, Department of Computer Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Iran;
2) Assistant Professor, Digital Processing and Machine Vision Research Center, Najafabad Branch, Islamic Azad University, Najafabad, Iran. Department of Electrical Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Iran;
3) MSc Student in Artificial Intelligence and Robotics, Department of Computer Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Iran;
4) MSc Student in Artificial Intelligence and Robotics, Department of Computer Engineering, Najafabad Branch, Islamic Azad University, Najafabad, Iran;

IOI Code : XCCE-EEAHB
Publication : 2nd International Congress On Engineering, Technology and Innovation(eticong.com/2nd)
Abstract :
    The present work studied about an off-line, text-independent approach based on both textural and structural features for identification and validation of the author s identity in a Persian handwritten text. Because there is no standard popular database for Persian handwritten words in order to experiment, we developed a new database of offline Persian handwriting text for the writer identification purpose. Textural features cannot extract those structural features, such as inter-word or inter-line spacing. For this purpose, this study aimed to increase the accuracy of the proposed system beside the new textural and structural features introduced through integrating them as a unique feature. In first step, the proposed method extracts three structural features, including the average inter-word and inter-line spacing, as well as simultaneously distribution of direction and area for the closed loops. Then, three extended features were extracted based on the local binary pattern and the textural features were achieved from the text. Each handwritten document is represented by an integrated triple histogram using the binary patterns in the image, and finally the integrated structural and textural features were prepared as the final feature to the nearest neighbor classifier in order to identify the author. The proposed method was evaluated on both Arabic and Persian datasets. Results suggested that the greater accuracy will be obtained using integrated textural and structural features consistently.

Keywords : author identification extended binary pattern structural features persian script