Paper Title
Handwritten Gujrati Script Recognition using Deep Neural Networks and OpenCV

Abstract
Sharing information in the form of text has always been predominant in the development of the human race. The traditional way of storing scripts on a piece of paper is susceptible to lose, damage and weather conditions. Thus the motivation behind this work is to develop an efficient mechanism for storing handwritten scripts digitally, to reduce the usage of paper in the process of documentation and to safely backup important data printed on paper by converting it into a text file. This mechanism involves extracting text from an image and converting it into a digitally editable format. Because of its unique features, Gujarati is the script used in this work. The main challenge in working with Gujarati script is the recognition of its characters because of their curves and the fact that the individual characters in the words are not connected. For the extraction of these characters from an image, thresholding is applied to the image, after which the image is sent to a clustering algorithm where individual characters are extracted and are resized. Once the clusters are generated, they are passed through a Deep Neural Network (DNN) for the classification of the characters. After this, the output labels of the DNN are associated with their respective Unicode and the result is printed in a text file. By following this approach, the mechanism solves the problem of storing and maintaining the printed document by being a complete framework for converting Gujarati scripts into a digital file with a current accuracy of 94%. This framework can also be used in industries with a lot of documentation expenditure, as this system aims to reduce paper usage in the documentation process. Additionally, this framework also involves a training module for further improving the efficiency and a module for easily generating training data from images. Keywords - Script Recognition, DNN Algorithm, Thresholding, OTSU, CNN, Support Vector Machine.