Prepoznavanje tabel in njihovih struktur iz dokumentov v formatu PDF in slik

Korbar, Bogdan (author), Žabkar, Jure (mentor)

V okviru razvoja programske opreme za upravljanje z dokumenti pri podjetju EBA d.o.o. Ljubljana je bila zaznana težava pri pridobivanju podatkov iz tabel v različnih skeniranih dokumentih. Do sedaj je bil za ta problem uporabljen lasten OCR model, ki pa ne dosega več želenih rezultatov. Zato je bila izvedena analiza obstoječih rešitev, pri čemer se je za najbolj primerno izkazal Microsoftov Table Transformer. To je model globokega učenja, zasnovan za zaznavanje objektov, ki se uporablja za prepoznavanje tabel v PDF-datotekah in slikah. V diplomski nalogi je bil Microsoftov Table Transformer preučen, prilagojen in testiran za uporabo v dokumentnem sistemu EBA DMS. Za učenje modela je bil uporabljen nabor 296 dokumentov, označen s pomočjo VIA (VGG Image Annotator). Za testiranje pa je bilo uporabljenih 50 dokumentov, pripravljenih v podjetju. Rezultati so pokazali, da je Table Transformer dosegel nekoliko nižjo natančnost pri prepoznavi tabel v primerjavi z obstoječim OCR sistemom, vendar je dosegel nekoliko višjo natančnost pri prepoznavi strukture tabel. Kljub temu obstoječi OCR model pri prepoznavi tabel še vedno nekoliko prekaša Table Transformer. Na podlagi teh ugotovitev je bilo sklenjeno, da se obstoječi OCR model zaenkrat obdrži, ob nadaljnjem raziskovanju in izboljševanju metod prepoznavanja tabel.

• Language: Slovenian
• Keywords: prepoznavanje tabel, pridobivanje podatkov, globoko učenje, Table Transformer, OCR, digitalno upravljanje z dokumenti, vrednotenje
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-161307
• COBISS.SI-ID: 211904259
• Publication date in RUL: 09.09.2024

KORBAR, Bogdan, 2024, Prepoznavanje tabel in njihovih struktur iz dokumentov v formatu PDF in slik [online]. Bachelor’s thesis. [Accessed 19 May 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=161307

Secondary language

• Language: English
• Title: Recognition of Tables and their structure from PDF documents and images

Abstract

As part of the development of document management software at EBA d.o.o. Ljubljana, an issue was identified in extracting data from tables in various scanned documents. Until now, a proprietary OCR model was used to address this problem, but it no longer achieves the desired results. Therefore, an analysis of existing solutions was conducted, and Microsoft’s Table Transformer was identified as the most suitable option. This is a deep learning model designed for object detection, used for recognizing tables in PDF files and images. In this thesis, Microsoft’s Table Transformer was studied, adapted, and tested for use in the EBA DMS document management system. A dataset of 296 documents, annotated using VIA (VGG Image Annotator), was used for training the model. For testing, 50 documents prepared by the company were used. The results showed that the Table Transformer achieved slightly lower accuracy in table recognition compared to the existing OCR system, but it achieved slightly higher accuracy in recognizing table structures. Nonetheless, the existing OCR model still slightly outperforms the Table Transformer in table recognition. Based on these findings, it was decided to retain the existing OCR model for now, while continuing to research and improve table recognition methods.

• Keywords: table recognition, data extraction, deep learning, Table Transformer, OCR, digital document management, evaluation