Exploring open source AI-OCR for efficient personnel invoice verification

Details

Supervisors

Schaus, Pierre ; Deville, Yves

Faculty

Ecole polytechnique de Louvain

Degree label

Master [60] en sciences informatiques

Abstract

This dissertation presents the development and evaluation of Docspotter, an open-source Python package designed to facilitate personnel invoice verification using Optical Character Recognition (OCR) technology. The motivation behind this study is rooted in the need to enhance the efficiency and accuracy of invoice processing at academic institutions, where traditional manual methods are labor-intensive and error-prone. Using a combination of CRAFT for text detection and Pytesseract for text extraction both prominent open-source tools—this work addresses the challenge of skewed text in diverse document formats that typically hinder OCR performance. Through the implementation of Docspotter, followed by a user-friendly GUI application, this work demonstrates the potential of leveraging deep learning and OCR techniques to streamline administrative processes. The evaluation conducted across various document types—scanned, pictures, and electronic—highlights the system’s precision and identifies areas for further improvement compared to industry standards, such as Azure AI Vision. The findings suggest that while Docspotter provides a robust platform for OCR tasks, incorporating adaptive processing techniques and advanced machine learning models could further enhance its accuracy and reliability. This work contributes to the field of document processing by demonstrating the practical application of open-source AI-OCR technologies in improving institutional administrative efficiency.