Skip to main content

OCR

Converts an image or a scanned PDF (support document) into usable text — optical character recognition — with many image preprocessing settings to improve reading quality.

Ports​

  • Input (left, image type) — typically downstream of a MIME Filter.
  • Output (right, support document type).

Engine​

  • Tesseract (default) — built in, free.
  • Configured OCR model — relies on an OCR model registered by a manager (for example Mistral OCR).

Tesseract settings​

  • Language: French + English (default), French only, or English only.
  • Page segmentation mode (PSM) — default 3 (automatic segmentation).
  • Advanced configuration (dedicated modal):
    • OCR engine (OEM) — default 3.
    • Resolution (DPI) — 72 to 1200, default 300.
    • Image preprocessing (checked by default), and if enabled:
      • Minimum upscale in pixels — 200 to 6000, default 1600.
      • Contrast mode — automatic (default), CLAHE, or none.
      • Binarization — none (default), Otsu, or adaptive.
      • Denoising, deskewing and sharpening — unchecked by default.
    • Raw Tesseract options as free text (e.g. --psm 11 -c preserve_interword_spaces=1).

"OCR model" settings​

  • Selection of the configured OCR model.
  • Target pages (e.g. 0,2,5-7).
  • Confidence score granularity (none / page / word).
  • Minimum confidence (0–100).
  • Free-text annotation instructions.

Both advanced configuration modes include a test panel: upload a sample image and run an actual OCR pass with the node's current settings, to calibrate them before running a real analysis (nothing is saved, it's just a preview).