Share the photo, not the person

Every week a picture is published with a face "blurred" that is still recognizable. The reason is almost always the same: the blur was applied with a radius too small for the size of the face. A gaussian blur averages each pixel with its neighbours; if the radius covers only a few pixels, the overall shape of the eyes, the nose and the mouth survives, and the human visual system is extremely good at recognizing a face from a shape alone.

The rule of thumb is to blur by at least a fifth of the face width. A 200-pixel face needs a radius of forty pixels, not five. At that strength, the colour and light of the scene remain — the picture still reads as a photograph, the person does not read as a person. Anything weaker is decoration rather than protection.

Why modern tools make weak blurs worse

De-blurring models trained to restore old photographs can partially invert a light blur, especially on a still subject with a clean background. They do not recover the original pixels, but they recover enough structure to make an identification plausible. Pixelation with large blocks removes the gradient information those models rely on, which is why it is the safer graphical choice, and a solid fill removes everything.

The deeper point is that anonymisation is a decision about consequences, not about aesthetics. If the picture protects a child, a patient, a witness or a source, the box should be unmistakable in the exported file. If it merely tidies a street scene, a soft blur is fine and looks better.

The details people forget

• Number plates: a car ties a place and a person together and is often left untouched.
• Screens: a laptop behind the subject can show a name, a chat or an inbox.
• Wristbands and badges: names, hospital numbers, barcodes.
• Reflections: windows and car paint hide a second view of the scene.
• House numbers and street signs, which locate someone precisely.
• Metadata: the photo itself may carry GPS coordinates even when every face is hidden.

Doing it in a browser, on purpose

Photos that need face blurring are, by definition, photos of people who have not consented to publication. Uploading them to a service to anonymise them is a contradiction: the service now holds the unblurred original. Running the blur locally removes that step. The picture is decoded into the canvas of your own tab, the boxes are drawn, the new file is written, and nothing is transmitted — which is also why the tool works with the network disconnected and why no usage log can contain the image.

A short procedure worth following

Open the largest copy you have, not a compressed thumbnail. Draw boxes that extend a little beyond the face — hair counts. Apply a strong blur or large-block pixelation, export as PNG, then open the exported file and look at it at 100% before posting. Finally, if the photo came from a phone, strip the location metadata with the EXIF remover: a hidden GPS point can undo every box you drew.

What the rules usually expect

Data protection rules in most countries treat a face as personal data, and publishing one without a basis is the problem — blurring is one way to remove that basis, but only if it works. Regulators have repeatedly said that reversible or weak anonymisation is not anonymisation at all, which is why the strength of the effect matters more than the tidiness of the result. For a school, a clinic or a workplace, the safe habit is to blur by default and ask before publishing an identifiable face.

Public places are not a free pass either. Someone walking past your shot has not agreed to appear in your marketing, and a blur removes the argument in one gesture. This is not legal advice, only a description of the direction the rules point.

Exporting without losing the work

Save the blurred copy under a new name and keep the original in a private folder, not in the same directory you upload from. Export PNG when the picture will be used on a page or in a document, because a second lossy compression pass softens the area around the boxes and can make a strong blur look uneven. If the destination demands JPG, use the highest quality the size allows.

And check the edges. A box drawn two pixels inside the face leaves a strip of ear or jaw that is often enough. Painting slightly wider costs nothing and is the difference between a picture that protects someone and one that merely looks as if it does.