Extracting handwriting and OCR text

This topic contains tips and tricks for extracting handwriting and OCR’d text, for example from scanned documents or images:

OCR tips

Sensible provides confidence scores for OCR’d text in the extraction when you configure high verbosity, so you know whether the extracted output comes from high- or low-quality text images. For document types that use OCR, write validations to warn you about extractions from low-quality scans.

Handwriting tips

• Choosing an OCR engine: Choose Google OCR. To configure OCR, click the gear icon for the Document Type and select Google:
  
• Defining regions: Handwriting can occupy an unpredictable region or even overlap other lines. To capture handwriting, Sensible recommends defining a region with a small height and long width that runs through the middle of the area that can contain the handwriting. The green boxes in the following image show this approach:
  
  For more information about how Sensible determines whether to extract a line that partially overlaps a region, see Region.
• Correcting for vertical misalignment: Jitter in the vertical positions of handwritten lines can cause Sensible to incorrectly sort lines that a human reader interprets as following left to right. The Sort Lines parameter corrects this problem by sorting lines by their likely reading order. For more information, see Sort Lines example.