Document Digitization Text Recognition
Loading...
Files
Date
2020-10-09T12:12:02Z
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This project highlights how Extended Modified National Institute of Standards and Technology (EMNIST) database can be put to use, to create clean and synthetic images of texts in different handwriting styles from the set of available images. This project presents a detailed review of Optical Character Recognition, the methods using which we can classify character by detecting and extracting the position of a character from synthetic images. With the help of the image processing, the raw image is cleaned by first converting the image into grayscale and applying the thresholding techniques, before it is sent for classification, so as to give a higher probability of successful recognition of characters. Once cleaned, the characters in the images are segmented based on lines, words and characters. These individual character images are resized to 28* 28 pixels and fed into the neural network for classification. The predicted string after classification is checked for any errors in spellings and finally, the output is written in an XML or JSON file which contains the predicted string, position of the string as well as the likely words.
Description
Keywords
Athira Ajay Kumar Kullully, Document Digitization Text Recognition, 1NH16EC012