An Useful Review on Optical Character Recognition for Smart Era Generation

Abstract

In this chapter, the authors have reviewed on optical character recognition. The study belongs to both typed characters and handwritten character recognition. Online and offline character recognition are two modes of data acquisition in the field of OCR and are also studied. As deep learning is the emerging machine learning method in the field of image processing, the authors have described the method and its application of earlier works. From the study of the recurrent neural network (RNN), a special class of deep neural network is proposed for the recognition purpose. Further, convolutional neural network (CNN) is combined with RNN to check its performance. For this piece of work, Odia numerals and characters are taken as input and well recognized. The efficacy of the proposed method is explained in the result section.

Design of Variational Autoencoder for Generation of Odia Handwritten Numerals in CNN Model

“…Dash et al have given a summarized study [9] on different methods used in both Odia printed and handwritten alphanumeric recognition where no work has been found to implement VAE. For Odia handwritten numeral recognition, a study has been conducted, a combined CNN-RNN model has been proposed in [10], and recurrent networks such as long short-term memory (LSTM) have been used by Das et al [11] and the accuracy in recognition in these approaches are found to be 99.99% and 97.93%, respectively. Application of support vector machine in Odia handwritten recognition has been found in the work of Sanjibani Sudha [12] where the model provided 85% accuracy.…”