This project provides a robust solution for improving Optical Character Recognition (OCR) accuracy, particularly for non-English languages or noisy inputs. It targets developers and researchers needing enhanced text extraction by intelligently combining multiple OCR engines with Large Language Models (LLMs) for correction and reconstruction.

How It Works

BetterOCR leverages a multi-engine approach, integrating EasyOCR, Tesseract, and Pororo (for Korean/English). It then utilizes OpenAI's chat models to refine and correct the combined OCR outputs. An optional custom context feature allows users to provide specific keywords or product names, significantly improving accuracy for specialized terminology and reducing errors.

Quick Start & Requirements

• Install via pip: pip install betterocr
• Requires Python. Pororo integration has additional dependencies. OpenAI API key is recommended for LLM functionality.
• See Examples for usage and performance.

Highlighted Details

• Combines EasyOCR, Tesseract, and Pororo for text detection and recognition.
• Employs OpenAI LLMs (GPT-3.5/GPT-4) for error correction and text reconstruction.
• Supports custom context for improved accuracy with specific terminology.
• Offers box detection functionality for locating text regions.
• Demonstrates improved results across English, Korean, and Hindi examples.

Maintenance & Community

• Project is under rapid development. Contributions are welcomed.
• Supported by (주)한국모바일상품권(Korea Mobile Git Card, Inc.).

Licensing & Compatibility

• Licensed under the MIT License.
• Permissive license suitable for commercial and closed-source use.

Limitations & Caveats

The package is under rapid development, with features like async support and an improved interface noted as "coming soon." Performance may vary based on OCR engine updates and OpenAI API availability.