Image Tools·4 min read

Blurring Sensitive Information: When Blur Fails You

That screenshot you're about to post has your account number in the corner. The reflex is to smudge it and hit send, and most days that's fine, but blurring sensitive information has a complicated track record: researchers keep finding ways to read what's underneath. This article explains what the Gaussian blur in Blur Image really does to pixels, how strong is strong enough, where blur genuinely fails as redaction, and what to reach for on the days it does.

Open the free Blur Image All insights

Why blurring sensitive information became the default habit

One slider, instant plausible privacy.

Screenshots of banking apps, support tickets with customer emails, ID photos sent for verification, bystanders in a street shot: all of them routinely get the blur treatment before sharing. Blur won that job because it's one control, it looks professional, and it doesn't scream redaction the way a black bar does.

Blur Image has three modes. Whole image blurs the full frame. Selected areas blurs only the rectangles you drag on the preview, so you can cover an account number, a face and a house number in one pass. Everything except areas does the reverse and keeps your rectangles sharp. Undo last and Clear areas fix a bad box, and the preview updates as you work.

What the blur radius actually controls

Radius is the whole story.

A Gaussian blur replaces every pixel with a weighted average of its neighborhood, and the radius, adjustable here from 1 to 50 pixels, sets how wide that neighborhood is. Crucially, the information isn't deleted. It's spread out. A digit blurred at radius 6 still contributes its ink to the surrounding smear in a mathematically predictable way. The tool pads the image with its own edge pixels before blurring, so borders keep their colour instead of fading dark.

Radius is measured in source pixels, so the same setting hits differently at different resolutions. Radius 12 on a 1200 pixel wide screenshot is a heavy smear; on a 4000 pixel photo it's a light haze. Judge by the preview, not by the number.

Is blur reversible? What the smear-recovery research shows

More often than you'd hope.

Because blurring is well-defined math, it can be partially inverted. Deconvolution sharpens lightly blurred images when the blur kernel is known or guessable. For text the attacks get uglier: an attacker who knows the font and layout, which is exactly the situation with bank statements, invoices, and plates, can render candidate strings, blur each one identically, and compare against your smear until one matches. Account numbers have a small character set, which makes the search cheap.

Heavy radii plus the re-encoding a canvas export applies do destroy a lot of that structure. But the safe mental model for anything high-stakes is that a light blur is decoration, not redaction.

A worked blur example on a fake bank statement

Watching digits die, radius by radius.

Take a 1920x1080 screenshot where the account number renders about 16 pixels tall. Choose Selected areas and drag a box over that line. At radius 4, the digits are still directly readable. At radius 12 they're gone to the eye, but the smear still carries structure a matching attack could exploit. At radius 30 the entire line is a featureless band of colour, and the download arrives as statement-blurred.png.

Faces follow the same curve at bigger numbers: recognizable at 8, ambiguous at 15, anonymous around 30 and up, always relative to how large the face is in the frame.

Blur mistakes that leak the thing you hid

Where careful people still slip.

The blur itself is rarely the weak point. The process around it is.

Choosing a radius under 10 for text and trusting that unreadable-to-me means unreadable. Matching attacks don't use eyes.
Redacting in software that layers the blur over the original, like some PDF and slide editors, where deleting the layer restores everything. A canvas export like this tool's contains only the blurred pixels.
Hiding the face but keeping the tattoo, the name badge, the reflection in the window, or the street sign behind them.
Posting the blurred copy while an unblurred version stays live in a cloud album or an old post.
Forgetting metadata entirely. Check the file with the EXIF Viewer, since pixels aren't the only channel that leaks.

How strong to blur, and two safer habits

Rules of thumb that hold up.

For text you genuinely need dead, start at radius 30 and confirm nothing survives when you zoom in. For batches, remember the mode, radius and any areas you drew on the first image apply to every file, with areas placed at the same relative position. That suits a set of identical screenshots, but check each result, and tune the radius against the largest image in the set, where it has the least effect.

And when the stakes are real, legal documents, other people's data, anything regulated, skip blur for the sensitive region and use an opaque block instead, such as the Redact tool in the Screenshot Editor. Blur earns its place for aesthetics and low-stakes courtesy hiding, not for secrets.

Blur or Pixelate Image? Choosing your censor

Two effects, different failure modes.

The Pixelate Image tool is the sibling to compare against. A mosaic reduces each region to flat averaged squares, which leaves less exploitable structure than a smooth smear at comparable coverage and visibly announces that redaction happened. Blur wins on looks, which is why it owns backgrounds and depth effects. Everything except areas gives a quick sharp-subject, soft-scene look with a rectangle; for a precise outline around a person, cut them out with the Background Remover and layer the two yourself.

A reasonable split: blur for atmosphere and mild courtesy, mosaic with large blocks for plates and faces, solid fills for anything you'd regret leaking.

Try it now

Open Blur Image

The tool is one click away. No sign up, no upload, no payment.

Open Blur Image

Write a review