Text recognition (OCR) for patient records digitization using CNN

Optical character recognition (OCR) is widely used to transcribe texts from images in computer vision. Although current OCR methods can accurately transcribe printed text (structured), they often fall short on unstructured or handwritten text recognition. This project proposed a text recognition met...

Full description

Bibliographic Details
Main Author: Ong, Zi Leong
Format: Final Year Project / Dissertation / Thesis
Published: 2022
Subjects:
Online Access:http://eprints.utar.edu.my/4662/
http://eprints.utar.edu.my/4662/1/fyp_CS_2022_OZL.pdf
Description
Summary:Optical character recognition (OCR) is widely used to transcribe texts from images in computer vision. Although current OCR methods can accurately transcribe printed text (structured), they often fall short on unstructured or handwritten text recognition. This project proposed a text recognition method to recognize handwritten text on patients' clinical data using a convolutional neural network (CNN). We compiled custom handwriting datasets from MNIST 0-9 and Kaggle A-Z datasets to add more handwriting diversity in training a more robust OCR model. The CNN has 3-convolutional layers to learn high-level features and a dropout layer to prevent overfitting. The preliminary results showed that the proposed model achieved 93.75% classification accuracy while Tesseract (the state-of-the-art OCR) scored 69.79%. The data will be transformed from handwritten text to computer-readable text and then stored in files in xml form for further development.