Comparative Evaluation of Text Recognition Methods for Field Extraction in Brazilian National Driver’s Licenses

With the advancement of digitalization, the demand for document information extraction solutions has grown, but studies focusing on Brazilian Driver’s Licenses (CNH) remain scarce. This paper highlights this gap by evaluating OCR methods, including TrOCR (with varying settings), Tesseract, and GPT models, for extracting information from CNH documents. The study considers challenges such as lighting and variable image quality, using metrics such as error rate and recognition evaluation in specific fields. The results demonstrate promising performance using open source tools, offering insights into the advantages and limitations of each model.