Use of Deep Learning in Digitization of Libraries

Author Name(s): Gajendra Muley, Mangesh Ambhure, Bhavesh Pawar, Simran Khiani
Author Email: nmuley0@gmail.com

Abstract

This is a proposed idea for the formation of digital databases, which contain information about all the books present in a library, without human intervention. We propose that deep learning is used to scan a book, extract its title and author and store the obtained information in a database which can later be accessed by the members of the library. An idea of how Hough Transform can be used to detect the boundaries of a book is elaborated. The main Deep Learning Techniques used are Convolutional Neural Networks, Recurrent Neural Networks. We have briefly explained the basic Artificial Intelligence concepts of Optical Character Recognition and Natural Language Processing. The complete mechanism of the proposed system and the problems or shortcomings of the system are discussed before the conclusion. We would like to conclude that by using the aforementioned practices we can develop a smart system to digitize book inventories and categorize the books on various fronts, without human intervention.

Introduction

A large number of literature sources are available in various libraries of India. These books are means of crucial information for various sectors. Due to a large number of books in a library the right book is often not found. This problem can be solved if we create a database of all the books present in a library and provide the user with the option to search through the database for a specific title or an Author of their choice. The formation of a digital database of thousands of books is very time and labor consuming. To reduce and minimize this effort, we can use Deep Learning and Artificial Intelligence techniques to identify the title of a book and store it in a digital database. To achieve this, we will use Optical Character Recognition (OCR). Most OCR algorithms perform poorly in low light or on distorted images. We will discuss techniques such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) that are used to recognize the text on book covers or the spines of books. These deep neural networks also help in improving the results of OCR. The support of techniques such Hough Transformation helps in early stages on text recognition and is also explained. We can use Natural Language Processing (NLP) and further query the obtained title on the internet and get information about the genre and price of the book. We establish that using deep neural networks and various deep learning algorithms we can detect the title of a book, its genre and locate it inside a library.

Conclusion

By using various deep learning techniques such as Hough Transform, Convolutional Neural Networks for OCR, Recurrent Neural Networks for NLP and connecting the obtained output to the internet we can accomplish the process of creating a digital database of the books present in a library and localize this information. This study shall be complete upon the solving the problem of inability of the system to differentiate between the title and author that might appear on the cover page of the book.

962 total views, no views today

Download PDF File

About the author: admin