Reconhecimento óptico de caracteres para leitura de documentos em formato PDF

Abstract

Given the inaccessibility of certain books and other texts published before the digital age, as well as the secondary treatment by the OCR community toward the processing of analyzed documents into digital text suitable for casual reading, the need is determined for a tool that, working with OCR, dedicates itself to the extraction of text from images or PDF files, aiming to facilitate the reading and dissemination of these texts. To achieve this objective, a comparative study was conducted between the OCR tools Paddle, Docling, and Tesseract, and the mLLM Gemini. A Python API was built to perform the OCR processing work and text formatting. This was integrated with a web interface developed using the PHP Laravel framework, with the aim of making the project available to the greatest number of people possible. The present work has both a scientific and experimental track, as well as a practical one, seeking to compare OCR tools and solve something that is often ignored by these tools: the proper formatting of the text for reading, also aiming to make the results of this research available as a tool that is easy for the community to use and access.