Mobile application for optical character recognition (OCR) of digits

• Main Author: Luah, Yee Teng
• Other Authors: Miao Chun Yan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/153186
• Institution: Nanyang Technological University

Technology has progressed tremendously throughout the years and humans have benefitted from this through its convenience and pervasiveness. As technology evolves, it will aid humans in ways that are unimaginable in the past. Every industry has been reaping the benefits of technology, including the medical field, where it has helped provide digital solutions to many healthcare problems. This project explores the utilisation of a built-in-camera in a mobile device together with an OCR model to function as the eyes of a human to recognise digits on health monitoring measuring devices. Such devices are used to monitor the health conditions such as weight, blood cholesterol, blood pressure and blood glucose of patients. Most existing solutions require manual input of the readings into the corresponding fill box of an application to track health conditions. However, this often leads to incorrect input of readings due to human errors. There are also research studies done applying Computer Vision models to aid inputting of readings. Nevertheless, the accuracy is not satisfactory due to the limited variety and quantity of data samples being used in their training. Therefore, this project aims to build a model using Convolutional Recurrent Neural Network (CRNN) tool package, together with over twenty thousand human annotated images of different devices. To further improve the recognition accuracy, an API is created for sanity checks based on the domain knowledge adopted from clinicians. With the model and API being applied, a user can just take a photo of a health monitoring device using a phone, and the application will recognise the digital readings autonomously without the requirement of manual inputs. Thus, the improved application reduces human error and greatly enhances usability.