Aplicación de EasyOCR para la extracción automática de información en documentos digitalizados: estudio de caso en Banco Davivienda

Abstract

This study analyzed the performance of EasyOCR for the automatic extraction of information from digitized documents. The study was developed based on document processing activities carried out during an internship at Banco Davivienda, where artificial intelligence tools were used to automate information extraction. For the evaluation, EasyOCR was implemented on a dataset of 32 digitized Colombian national identity cards. Image preprocessing techniques, including orientation correction and Gaussian filtering, were applied. Subsequently, accuracy metrics were calculated for different information fields contained in the documents. The results show that the configuration combining GPU processing and image preprocessing achieved an overall accuracy of 90.31%, outperforming the configuration without preprocessing. Additionally, the study found that image quality directly affects system performance, particularly for fields containing larger amounts of text. Finally, the study provided an understanding of the main components of the CNN+CTC architecture used by EasyOCR and demonstrated the potential of these types of tools to support automated document information extraction processes.