Automatic Information Extraction From Student ID Card Images Using DB and VietOCR: A Case Study at a Vietnamese University

The development of an information extraction system from student ID card images plays an important role in the digitalization of student management. This study proposes a two-stage processing framework that integrates computer vision and deep learning techniques, in which MobileNetV3-Small is employed for student identification card image classification, while the Differentiable Binarization (DB) model and VietOCR are responsible for Vietnamese text detection and recognition, respectively. Experimental results on a student ID card image dataset show that the classification model achieves an accuracy of 99.40% with an AUC of 0.9996, while the DB-based text detection model attains an Hmean of 89.81% after data augmentation. For text recognition, the proposed system achieves over 99% character-level accuracy and up to 98.90% full-sequence accuracy. These results demonstrate the effectiveness and practical feasibility of the proposed system, which is further validated through a proof-of-concept offline attendance application. In addition, the system is designed with computational efficiency in mind, enabling deployment on resource-constrained devices without requiring continuous internet connectivity. The proposed framework can be readily adapted to other types of identification documents, providing a scalable and cost-effective solution for automated data acquisition in educational institutions.