Redact scanned documents safely: a box for every word
A black bar over an account number. Looks done. Then someone selects the bar, copies, and the number is in the mail. Which means: the bar belongs on the box of each word, not on a picture of one.
Try it on one of your own pages and see the box of every word, the places a redaction bar has to cover.
Word One word, Northlight. Box measured on the image.
Redacting a scan: three ways a black bar fails
- A black bar over the text, while the text layer underneath is still there and can be copied.
- A word the OCR engine missed, so no bar covers it.
- A box a little too small, so the edge of a letter shows.
OCR for redaction: what the result delivers
A box and a confidence for every word
With a pipeline → that returns a box for every word.
A status for every page
A page that could not be read is named in the header X-Pages-Failed: you know where nothing was checked.
How to redact a scanned document in six steps
- Send the page and ask for
format=json. - Search the words for your patterns: account number, name list.
- Draw the boxes of the hits, with a margin.
- Burn the bars into the pixels of a new image, not a layer over the old one.
- Bar words with a low confidence next to a hit too, or have a person check them.
- If a page has the status failed, hold it back.
Finding the personal data is your part: your patterns and name lists run on the words and their places.
How exact are word boxes for redaction
Boxes are a starting point: you set the margin and the check of unsure words.
Six clean digital test pages, own run. The truth is the text layer of the PDF.
Account numbers and names stay in your own network when the redaction runs on your own server.
Try OCR for redaction free
Read one of your own pages free and check the box of each word, or open the OCR demo gallery to see the box a black bar would cover.
Production use, high volume or your own server?
Private use and tests are free. For a company in production, a server of your own or a pipeline built on your pages, send the kind of page and the volume, and get a written offer, or a clear no when the fit is not there.
