Evaluate OCR Models and Finalize Best Fit

[SahilBheke25]

Evaluate OCR Models and Finalize Best Fit (Refer: Wiki Report)

Description

We have tested multiple OCR models to evaluate their accuracy, ease of use, and performance across English and Marathi text inputs.

The detailed results and observations are documented in the project wiki:

🔗 OCR Model Comparison Report

---

Summary of Findings

• Tesseract
  Best overall performance. Accurate, lightweight, and easy to integrate.

• EasyOCR
  Moderate performance; issues with structure and sentence clarity.

• I2L-NOPOOL & DTrOCR
  Requires additional training or lacks proper documentation.

• GPT-4o
  Requires GPT Plus subscription (not free).

---

Additional Notes

• This issue is linked to the OCR research wiki for ongoing reference.
• We may consider post-processing like word segmentation or spell correction in the future for further improvements.