EXTRACTING INFORMATION FROM REGISTRATION FORM FOR FOREIGN LANGUAGE PROFICIENCY EXAMINATIONS

Abstract

This thesis focuses on overcoming the challenge of extracting candidate information from examination registration form, especially in contexts like competitive exams where processing large volumes of such forms efficiently is crucial. It proposes advanced Optical Character Recognition (OCR) techniques customized for the unique demands of examination data processing. In the topic: "Extracting information from foreign language proficiency test registration form" in addition to knowledge about image processing, the topic also focuses on research on word recognition using the CNN model combined with Transformer architecture and Training steps to increase recognition ability. In this project, we developed a lightweight OCR model specifically tailored for recognizing Vietnamese words and handwriting. Our model effectively decodes characters with high accuracy and speed, utilizing a blend of CNN, Transformer, and cross-entropy loss. These enhancements not only address the challenges of Vietnamese handwritten OCR but also hold promise for broader applications in image processing and computer vision.