An Approach for OCR detection and Classification for Devanagari Printed Text using Deep Learning

Abstract

Optical Character Recognition (OCR) is the method of interpreting text from digital documents automatically. It is a large area of signal processing science. In several languages of India, such as Hindi, Nepali, Marathi, Sindhi etc., Marathi has used. The Marathi language is used by more than 300 million people worldwide. This pattern forms the basis of the communities of India. It plays a significant role in the production of manuscripts and literature. OCR research has been done for in banks, post offices, defence organizations, library modernization, etc., because of its potential applications. Several techniques are available for the character segmentation of handprint Gujrati, Bangla, Tamil, Hindi, etc., with these methodologies. However, much work is done for both the given material, but for the laboratories, it is only limited.. In this paper, proposed a Deep Learning based Convolutional Neural Network (CNN) classifier technique has used for OCR System of printed as well as scanned newsprint Marathi script. This research system deals with various feature extraction and feature selection techniques with CNN classification. In experimental analysis we train system around more than 51 characters that produces better detection accuracy for test images