Vision and OCR

How it works

Document extraction reads text and structure while preserving its source and correction history. OCR output needs review before becoming a trusted source.

Before you start

Use an authorized, supported file and preserve the original. For scans, check orientation, contrast, language and whether tables or handwriting require separate review before extraction.

What to do

1. Select an authorized document and an available extraction capability.
2. Wait for processing and inspect the extracted text.
3. Correct names, numbers, table structure and uncertain characters.
4. Save approved material to the intended Knowledge scope.

Verify the result

Poor image quality or complex layouts can produce errors. Compare uncertain passages to the original rather than treating extracted text as exact.

If it does not work

If extraction is incomplete, inspect the page or field evidence, improve scan quality, split an oversized document or upload a supported source format. Do not publish extracted values to Knowledge until uncertain fields and tables have been reviewed.