QuickieTool

How to blur faces and details in a photo before you share it

By Muhammad Zaheer, QuickieTool6 min read

You want to post a photograph but something in it should not be public: a child's face, a number plate, a house number, an address on an envelope, a screen full of somebody's account details.

Obscuring it is straightforward. Obscuring it in a way that actually holds up takes slightly more thought, and the reasoning is the same one that makes black rectangles a bad way to redact a document.

Hiding versus destroying

There is a meaningful difference between an edit that conceals detail and one that removes it.

A light blur averages each pixel with its neighbours. The original values are gone, but a lot of structure survives, and for a small set of possibilities — six digits on a number plate, four digits on a door — the surviving structure can be enough to narrow the answer considerably. The same is true of gentle pixelation across a small area.

A heavy blur or coarse pixelation genuinely destroys the information. When an entire region collapses into a handful of flat blocks, there is nothing left to work backwards from, because many different originals would produce that same output.

The practical rule is to be far more aggressive than looks necessary. If you can still see the shape of the digits, the edit is decorative.

  • Light blur over small text: often recoverable in principle, and always guessable
  • Heavy blur or coarse pixelation over a generous area: sound
  • A solid block of colour: the most reliable option of all
  • Emoji or a sticker placed on top: fine, provided it is flattened into the image
  • Any of the above applied in a layered editor and saved as a project file: not applied at all

That last one catches people. An editor that keeps layers keeps the original underneath. Export to a flat image format so the obscured region is genuinely part of the picture rather than something sitting above it.

Cover more than you think you need

Obscure a generous area around the detail rather than tracing its outline. A blur that exactly matches the shape of a number plate tells the viewer where the plate is and how large the characters were, and any edge that was missed becomes disproportionately informative.

For faces, cover the whole head rather than the features. A blurred face with a visible hairline, ear shape and jaw outline is still recognisable to anybody who knows the person, which is usually exactly the audience you were worried about.

Blur or pixelate part of an image before you share it. Runs in your browser, so the photo never leaves your device.

Blur part of an image

What else in the photo identifies you

Concentrating on the obvious detail while ignoring the rest of the frame is the commonest failure, and it has nothing to do with blurring technique.

  1. Reflections: windows, mirrors, screens, glasses and even a shiny kettle can show the room or the photographer.
  2. Documents in the background, including anything on a screen or a noticeboard.
  3. Distinctive surroundings: a street sign, a shopfront, a view from a window, an unusual building opposite.
  4. Uniforms, badges and branded clothing that identify an employer or a school.
  5. Location metadata stored inside the image file, which no amount of blurring touches.

That last one is invisible and outlives every edit you make to the picture itself. A photograph taken indoors typically carries GPS coordinates accurate to a few metres, and stripping them is a separate operation from blurring.

A sensible order of work

Crop first, so anything at the edges you did not intend to include is gone rather than obscured. Then blur or block out what remains. Then flatten and export. Then remove the metadata from the exported file.

Doing metadata last matters, because editors frequently write new metadata as they save. Stripping it before your final export means stripping something that gets recreated a moment later.

Assume anything you obscure will be examined by someone who wants to know what it said. Edit as though that is already true.