Optical character recognition using deep learning for keyword-triggered value extraction in documents

• Main Author: Lie, Valencia
• Other Authors: Lee Bu Sung, Francis
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/174994
• Institution: Nanyang Technological University

The manual labour hours needed to compare values within documents remain one of the top inefficiencies that Manufacturing companies like SLB face when performing their regular Quality Control Inspection. On top of that, the risks of errors are high due to its manual handling nature. To combat these problems, this research project aims to use automation and Machine Learning to extract meaningful key-value pairs within documents and compare them automatically. This is done in three different sections: extraction of texts from images using OCR engines, extraction of key-value pairs using Layout-based models and the linking of key-value pairs using Graph-based models and Proximity-based algorithm. On top of these three segments, a prototype is also developed to showcase the modules working hand-in-hand. Although there are past research projects that aim to tackle similar issues, most of the research projects only focus on one of the aspects mentioned above, instead of tackling the problem end-to-end. Furthermore, limited attention has been given to the manufacturing industry in this specific domain as other research projects mainly focus on documents from other industries, such as healthcare documents and retail receipts. While not fully commercially ready, the findings and development detailed in this research project impart valuable knowledge on how to tackle the issue at hand as it addresses multiple facets of the problem.