Review Paper on Marathi Handwritten Classification Using Deep Learning

Review Paper on Marathi Handwritten Classification Using Deep Learning

Prof. Abhay Gaidhani

(Project Guide)

Sandip Institute of Technology and Research Centre

Savitribai Phule Pune University

Nashik, Maharashtra, India

Rohit Loharkar, Tushar Sonawane, Rohan Adhav, Raj Patil

Sandip Institute of Technology and Research Centre

Savitribai Phule Pune University

Nashik, Maharashtra, India

Abstract— The most difficult task in today's world is handwritten character recognition in Marathi. It takes a lot of time and effort to share tangible documents. The proposed effort will reduce the demand for storage space, ease the difficulty of data entry in Marathi-language forms, and transform deteriorated historical records into editable text, which is a requirement for OCR to convert Marathi written texts or letters to editable text. Furthermore, because of their structure, shape, multiple strokes, and various writing styles, handwritten Marathi characters are frequently more difficult to read and understand. In this study, a single handwritten Marathi character is taken as input, and the zonal feature extraction technique is used to extract the features and characters are categorized using a deep convolutional neural network (DCNN). This recognition system mainly has five stages i.e., Data acquisition, Pre-processing, Feature Extraction, Character classification and recognition. Finally, we discuss some existing issues in character recognition and suggest future research and development areas that will lead to the usage of deep learning-based character recognition in applicable sectors.

Keywords—CNN, Deep Learning, Neural Network, Marathi character recognition, Zoning feature extraction.