Why blurring text is not redaction

Blurring a number in a screenshot feels safe because you can no longer read it. Whether a computer can is a different question.

Why blur can be undone

A blur does not delete information. It spreads each letter’s ink over its neighbours in a fixed, repeatable way. Different letters still leave different smudges.

If someone knows roughly what was hidden (say, digits in a known typeface), they can blur every candidate the same way and compare the results with your smudge. For a card number or a short password there are few enough candidates that this works quickly. Free tools exist that do exactly this.

Why pixelation is no better

Pixelation replaces each block of the picture with its average colour. Text is so regular that the pattern of averages is close to unique for each character. Researchers and hobbyists have repeatedly recovered text from pixelated screenshots, and a moving video makes it easier still, because each frame leaks a little more.

Faces are a different case

A blurred face gives a person looking at it very little, and that is usually the goal. It is not proof against recognition software, but for hiding a bystander in a photo it is a reasonable choice. For anything written, it is not.

The black box that is not one

Three ways a “black box” fails:

• It is a layer on top. In a PDF or a document, a drawn rectangle sits above the text. Select all, copy, paste, and the text comes out.
• It is semi-transparent. A highlighter set to black, or a marker tool in a phone’s photo editor, often leaves the pixels underneath slightly visible. Raising the brightness brings the text back.
• The original is still in the file. Some apps keep the untouched picture for “undo”, or leave a thumbnail of the original inside the file.

What works

Replace the pixels. A fully opaque, solid box that overwrites the area leaves nothing to recover, because the information is no longer in the picture.

• For a picture, Redact image paints solid boxes into the pixels themselves and writes a new file with no thumbnail or history. It offers blur and pixelation only with a warning, for faces.
• For a PDF, Redact PDF turns each redacted page into a picture with the boxes burned in, so the text, pictures and drawings underneath are gone, not covered. It then extracts the text of the result and confirms that none of the redacted words remain.

Two habits worth having

1. Cover a little more than the text. The length of a box says how long the word was.
2. Check your own work. Try to select and copy from the redacted file, and search it for the thing you hid.