About automatic recognition of printed text

Abstract

This article examines the internal structure of a system for recognizing printed data and converting it into a convenient format for use. The developed project provides a solution for converting images and PDF files containing printed text and formulas into LaTeX format by using free, open-source libraries and a custom-trained model for data classification, detection and segmentation. This study provides a detailed overview of the system development stages starting from the analysis of existing algorithms to the creation of a custom model for text and formula recognition. The focus is on selecting tools and libraries suitable for tasks related to the automation of recognizing and converting printed documents into LaTeX format and on attempts to improve the performance of these libraries. The process of creating and preparing the dataset for model training which includs image and text annotation. The outcomes of the model training are presented in tables showing the achieved recognition accuracies for various data classes. The article presents the results of testing the system's functionality for recognizing text and formulas in both Russian and English. It also details the strengths and weaknesses of the developed system and the challenges encountered during its development. Based on our development, we have created a website that includes an interface for converting images and PDF files into LaTeX format, as well as a Telegram bot with similar functionality