I.J. Information Technology and Computer Science, 2019, 6, 9-17 Published Online June 2019 in MECS (http://www.mecs-press.org/) DOI: 10.5815/ijitcs.2019.06.02 Copyright © 2019 MECS I.J. Information Technology and Computer Science, 2019, 6, 9-17 Off-line Sindhi Handwritten Character Identification Arsha Kumari Department of Electronic Engineering, Mehran University of Engineering and Technology Jamshoro, Sindh Pakistan E-mail: arsharathi56@gmail.com Din Muhammad Sangrasi*, Sania Bhatti*, Bhawani Shankar Chowdhry** and Sapna Kumari* *Department of Software Engineering Mehran UET Jamshoro, Sindh Pakistan **Meritorious Professor, Faculty of Electrical, Electronics and Computer Engineering, MUET Jamshoro, Sindh E-mail: din.muhammad@faculty.muet.edu.pk, sania.bhatti@faculty.muet.edu.pk, c.bhawani@ieee.org, rathisapna65@gmail.com Received: 26 March 2019; Accepted: 19 May 2019; Published: 08 June 2019 Abstract—Handwritten Identification is an ability of the computer to receive and translate the intelligible handwritten text into machine-editable text. It is classified into two types based on the way input is given namely: off-line and online. In Off-line handwritten recognition, the input is given in the form of the image while in online input is entered on a touch screen device. The research on off-line and online handwritten Sindhi character identification is on its very initial stage in comparison to other languages. Sindhi is one of the subcontinent's oldest languages with extensive literature and rich culture. Therefore, this paper aims to identify off-line Sindhi handwritten characters. In the proposed work, major steps involve in characters identification are training and testing of the system. Training is performed using a feed-forward neural network based on the efficient accelerative technique, the Back Propagation (BP) learning algorithm with momentum term and adaptive learning rate. The dataset of 304 Sindhi handwritten characters is collected from 16 different Sindhi writers, each with 19 characters. The novelty of proposed work is the comparison of the recognition rate for the single character, two characters and three characters at a time. Results showed that the recognition rate achieved for a single character is more than the recognition rate of multiple characters at a time. Index Terms—Off-line Handwritten, Neural Network training, Back Propagation (BP) algorithm, Sindhi Character identification. I. INTRODUCTION Handwritten character recognition (HCR) is advancing the communication between human and computer; it takes the world toward automation [1]. Off-line handwritten recognition is a somehow easy and fast way of inputting data to the computer. As plenty of Sindhi literature is available in Sindh literature departments in hard form, has taken too much space and will take too much time to access any information. Therefore it is necessary to preserve that information in the digitized form so that globally everyone can access easily. Hence Sindhi HCR is a very initial step to preserve the Sindhi literature on the web to use it at the worldwide resource. Handwritten recognition is a very challenging task in computer vision and pattern recognition since every writer has a different writing style, different shape of characters and font, image quality [2]. As Sindhi is cursive language in which characters are connected to form words, hence it is a more difficult job when it comes to recognize the off-line Sindhi handwritten characters. Another problem in recognition of Sindhi characters is similarity in 1) basic shape of characters, 2) position of dots and 3) the number of dots of different character [3]. Though much work has been done on other languages such as English [4], Chinese [5], Arabic [6] and other languages but very less work has been done on Sindhi HCR at the best of the knowledge, so a lot of work is required to be done in this direction. Since Sindhi is the regional and provincial language of Pakistan spoken by 60 million people in Sindh and different areas of the world [7]. Basically, HCR is classified into two types first one is off-line and another is online. Both these types vary from one another by the way the input is given to the system. In the off-line handwritten recognition, the input is given in the form of a paper document, image etc. that will be static in nature. While in online the input is given on the touch screen device such as Tablet etc., that input will be dynamic in nature. The proposed work main purpose is to identify the off- line handwritten Sindhi characters using the BP algorithm with adaptive learning and momentum which reduce the training time of the network. An additional contribution of this research is to perform the comparative analysis in the recognition rate for a single character and multiple characters at a time. This system is based on the graphical user interface (GUI) which is developed using MATLAB 2017a programming environment by utilizing its