nirmalpratheep/TamilTextCorpus

Project Madurai Corpus Notes

Overview

• This corpus consists of OCR text extracted exclusively from the https://www.projectmadurai.org/index.html landing page, respecting the distribution guidance shared by Project Madurai volunteers.[^1]
• The material supports Tamil language modeling experiments within this repository; please retain this README whenever redistributing the derived text files so that attribution remains intact.

OCR Pipeline

The corpus was generated with the Python workflow below. It renders each PDF page as a 300 DPI image, runs Tesseract with the Tamil language pack, and concatenates the recognized text into ocr_tamil_output.txt.

from pdf2image import convert_from_path
import pytesseract
import os

tesseract_cmd = r"C:\\Program Files\\Tesseract-OCR\\tesseract.exe"
if os.getenv("TESSERACT_CMD"):
    tesseract_cmd = os.getenv("TESSERACT_CMD")
pytesseract.pytesseract.tesseract_cmd = tesseract_cmd

tessdata_dir = r"C:\\Program Files\\Tesseract-OCR\\tessdata"
if os.getenv("TESSDATA_PREFIX"):
    tessdata_dir = os.getenv("TESSDATA_PREFIX")
else:
    os.environ["TESSDATA_PREFIX"] = tessdata_dir

tam_data_file = os.path.join(tessdata_dir, "tam.traineddata")
if not os.path.exists(tam_data_file):
    print(f"Warning: Tamil language data not found at {tam_data_file}")
    print("Please download tam.traineddata and place it in the tessdata directory.")

poppler_path = r"C:\\Popler\\poppler-25.07.0\\Library\\bin"
if os.getenv("POPPLER_PATH"):
    poppler_path = os.getenv("POPPLER_PATH")

pdf_path = r"corpus_data/project_madurai/pm0001.pdf"
output_txt = "ocr_tamil_output.txt"

pages = convert_from_path(pdf_path, dpi=300, poppler_path=poppler_path)

with open(output_txt, "w", encoding="utf-8") as out_file:
    for i, page in enumerate(pages):
        print(f"OCR on page {i+1}...")
        text = pytesseract.image_to_string(page, lang="tam")
        out_file.write(text + "\n\n")

print(f"OCR completed. Output saved to {output_txt}")

Key Configuration Notes

• tesseract_cmd and TESSDATA_PREFIX can be overridden via environment variables when running on systems with custom installations.
• Ensure tam.traineddata is present in the tessdata directory; install it from the official Tesseract repository if missing.
• Poppler binaries are required for pdf2image. Update poppler_path to match your local installation before running the script.

Acknowledgements

• Project Madurai retains all rights to the original digitized works. Please credit the initiative and preserve their logo whenever reusing this corpus.[^1]
• Logo reproduced here under the attribution guidance published by Project Madurai volunteers.[^1]

[^1]: Project Madurai, “Project Madurai,” https://www.projectmadurai.org/index.html (accessed 9 Nov 2025).