An Investigation for Cursive Context-Specific Printed Script Recognition Techniques

Selecting suitable techniques and datasets for a machine learning problem is a great challenge. This research investigates the available machine learning architectures, datasets, and software tools, for cursive context-specific printed text. Despite the existence of millions of speakers around the globe and rich literary history of more than thousands of years, it is hard to find computational linguistic works related to the Punjabi Shahmukhi script, a member of Perso-Arabic context-specific script – a low-resource language family. An investigation of related work through collected statistics inspired by the popularity and success of artificial neural networks and associated techniques have presented. The historical development of the script follows it in a literary sense and the importance and motivation in the development of optical character recognition systems, algorithms, and techniques. The paper discussed the available datasets and the need for a custom dataset. Based on recent research trends in machine learning frameworks, the work presents the most popular deep learning techniques, convolutional neural networks, and recursive neural networks for cursive context-specific scripts. This paper incorporates actual knowledge to promote research in machine learning and natural language processing disciplines and the selection of algorithms, architectures, and resources.