llm_aided_ocr

What it is

The LLM-Aided OCR Project enhances OCR outputs by applying LLM-based error correction and optional Markdown formatting to OCR text derived from PDFs via Tesseract. It supports both local LLMs and cloud-based API providers, with asynchronous processing for improved performance and detailed logging.

How it works

• PDF to image conversion using pdf2image
• OCR using Tesseract (pytesseract) with preprocessing steps
• Splits raw OCR into chunks for processing with overlap to preserve context
• For each chunk, uses an LLM to correct OCR errors and optionally format as Markdown
• Removes duplicate content during Markdown formatting
• Optional suppression of headers, footers, and page numbers
• LLM integration supports local LLMs (llama_cpp) and API providers (OpenAI, Anthropic), with error handling and token management
• Asynchronous processing via asyncio when using API-based LLMs to process chunks concurrently while preserving order
• Quality assessment comparing original OCR to processed output
• Logging throughout the process

Getting started

• Prerequisites include Python 3.12+, Tesseract OCR, pdf2image, PyTesseract, and optionally OpenAI/Anthropic APIs or a local LLM
• Installation sequence:
  1. Install Pyenv and Python 3.12 (provided script)
  2. Set up the project:

     git clone https://github.com/Dicklesworthstone/llm_aided_ocr    
     cd llm_aided_ocr          
     pyenv local 3.12
     python -m venv venv
     source venv/bin/activate
     python -m pip install --upgrade pip
     python -m pip install wheel
     python -m pip install --upgrade setuptools wheel
     pip install -r requirements.txt
     

  3. Install Tesseract OCR (system package dependent)
  4. Create a .env with API keys and settings, e.g.

     USE_LOCAL_LLM=False
     API_PROVIDER=OPENAI
     OPENAI_API_KEY=your_openai_api_key
     ANTHROPIC_API_KEY=your_anthropic_api_key
     

• Usage:
  1. Place your PDF in the project directory
  2. Update input_pdf_file_path in main()
  3. Run the script:

     python llm_aided_ocr.py
     

• Outputs include raw OCR output and final LLM-corrected Markdown or text files

Recent releases

• No releases listed in the latest entry

Traction

• Stars: 2972

Behind the repo

• Repository owner: Dicklesworthstone
• Language: Python
• License: none listed

Caveats

• License not listed in the repository data
• Requires external dependencies (Tesseract, pdf2image, OpenAI/Anthropic or local LLMs) and environment configuration
• Age: Created 2023-07-26; last push 2026-08-03