Optical Character Recognition for Marathi Language using Deep Convolutional Neural Network


Dr. Suresh Limkar, Gautami Mudaliar, Sneha Kulkarni, Neha Rathod, Tejasvi Gadakh, Sanaya Shah
Department of Computer Engineering, AISSMS IOIT, Pune, India.


Optical Character Recognition (OCR) is currently a developing idea provid-ing variety of advantages under its domain. It is the most basic stage of any document analysis system. OCR systems for English are available in abundance. But, applica-tion of OCR in Devanagari script isn’t so much of an explored portion. In a country like India, where most of the legal ancient documents are printed in the Devanagari script, the requirement is massive so it is important to create a commercially available software solution. This paper proposes a novel framework to recognize printed Mara-thi characters. It uses Deep Convolution Neural Networks. This Deep CNN algorithm will result in increased accuracy for pattern matching and overcome classification problems. The self-made dataset covers most of the characters present that makes it accurate predictions even more efficient. It also increases processing speed of charac-ter recognition and helps to compress the size of storage of the record, thus saving space.