OCR-RenAIssance

An OCR pipeline project for extracting main text from Renaissance-era scanned documents while ignoring embellishments. This project leverages EasyOCR (based on a CRNN architecture) to convert PDF pages to images and extract text efficiently.

Table of Contents

• Introduction
• Features
• Installation
• Usage
• Output
• Evaluation Metrics
• Contributing
• License

Introduction

The OCR-RenAIssance project focuses on developing an Optical Character Recognition system for Renaissance-era documents. The goal is to accurately extract the main textual content while filtering out any embellishments. This is crucial for historical document digitization and analysis.

Features

• PDF to Image Conversion: Utilizes pdf2image to convert scanned documents into images.
• Text Extraction: Implements EasyOCR for efficient text recognition.
• Evaluation Metrics: Calculates Character Error Rate (CER) and Word Error Rate (WER) to assess OCR accuracy.
• Clean Output: Outputs results in a well-structured text file with clear page delineations.

Installation

1. Clone the Repository:

   git clone https://github.com/icvasu/OCR-RenAIssance.git
   cd OCR-RenAIssance
   

2.Create a Virtual Environment

python -m venv venv
source venv/bin/activate   # On Windows: venv\Scripts\activate

3.Install Dependencies:

pip install -r requirements.txt

4.Install Poppler:

Windows Users: Download Poppler from Poppler for Windows, extract it (e.g., to C:\poppler), and add C:\poppler\Library\bin to your system PATH.

Mac/Linux Users: Install via Homebrew (brew install poppler) or your package manager. Usage 1.Prepare Your PDF:

Place your Renaissance PDF in the project folder.

Update the pdf_path variable in main.py with the correct path.

2.Run the OCR Script:

python main.py

3.Check the Output:

The OCR results will be saved in ocr_output.txt inside the repository.

Outputs are also organized in the outputs/ folder.

Output The ocr_output.txt file includes the extracted text from each page, clearly separated by page numbers.

Evaluation Metrics To assess the performance of the OCR model, the following metrics are used:

Character Error Rate (CER): Measures the accuracy at the character level.

Word Error Rate (WER): Measures the accuracy at the word level.

These metrics help in fine-tuning the OCR pipeline and ensuring reliable performance.