Research into merging Tesseract with LLMs [google gemini] - MadMouse/FuelBeacon-ML GitHub Wiki

Typical OCR text extraction -

• Pre-processing
  • Noise reduction
  • Image Binarization, have been using greyscale
  • Deskewing
• Text segmentation
• Character recognition
  • Pattern matching
  • Feature extraction
  • Machine learning algorithms
• Post-processing
  • Spell-checking
  • Language-specific rules

Gemini text extraction -

• Create prompt to be given to model, instructing fields to be extracted, returned in json format
• Provide image and prompt
• Convert json into pandas data frame? Think jsons can be sent directly to backend?

Gradio - opensource python library, rapidly build and share machine learning demos and applications [gemini_gradio - github https://github.com/AK391/gemini-gradio]

Gradio 'Image' - [gradio.Image()] upload images as input or display as output