[1Yin-A] Poster 1

[1Yin-A-24] Information Extraction and Structuring from Residence Card Document Images Using Deep Learning-based OCR

〇Chie Ito¹, Yijun Feng¹ (1. Daiichi Institute of Technology, Faculty of Engineering, Department of Information, AI and Data Science)

Keywords:

Deep Learning-based OCR,Document Image Understanding,Information Extraction,ID Document Analysis

In this study, we designed and evaluated an information processing method that automatically extracts key information from residence card document images using deep learning–based OCR and outputs it as structured data. In recent years, the increasing number of international students has created a strong demand for improving the efficiency of residence card verification tasks at universities. To achieve flexible information extraction without relying on fixed templates, we constructed a processing pipeline that combines deep learning–based OCR with semantic reasoning using a Vision-Language Model. The extracted results are further processed using regular expressions and rule-based normalization to generate structured data in CSV/JSON formats. Experiments conducted on 20 scanned images and 11 smartphone-captured images demonstrated high extraction accuracy of 90–100% for seven items, including residence card number, name, and date of birth. For the address field, detailed analysis using the Character Error Rate (CER) confirmed that a certain level of recognition performance was achieved at the character level. These results indicate that the proposed method is useful as an information extraction framework for identity documents that can flexibly adapt to institutional changes and layout variations.