Cover a face in a photograph

Most requests to cover a face are not about the photographer’s own privacy. They are about somebody else: the four strangers at the tables behind a holiday snap, the other parents in a sports-day picture, a colleague who happens to be in the frame of a photograph about something else entirely. Those people had no say in the picture being taken and none at all in it being published.

That is why this route opens with an ellipse and with the solid fill. An ellipse because a rectangle over a head has four corners that announce exactly where the covering ends and how big the head underneath it was. The solid fill because the failure mode here is other people’s, and it is the only covering on this site that genuinely cannot be undone.

A blurred face in a group photograph is still a face-shaped smear of the right size, in the right place, with the right hair. Against a small set of candidates — a class, a team, a department — that is often enough.

Pressing the finder downloads a 224 KB model called YuNet from this network’s own object store, once, and then runs it on your processor in roughly sixteen milliseconds per picture. Nothing about your photograph is sent anywhere to make that happen; the model travels to you and stays on this device.

• The model reports a box and five points per face. The box is what this tool uses; the points only help it decide how far past the box a head extends.
• Every box is grown: a third of its width to each side, nearly half its height upward, a fifth downward.
• Every proposal arrives selected and covered, and every one can be moved, resized or deleted.
• A picture with no faces in it reports no faces rather than guessing, and says so in words above the canvas.

If the download fails, the interface says so plainly and the manual path is unaffected. A region you draw by hand covers exactly the same pixels in exactly the same way — the finder saves you time, not protection.

Open the saved file and look at it, rather than looking at the preview. It takes ten seconds and it catches the one failure this tool cannot catch for you: a region that was moved after it was drawn, or a face at the edge of the frame that nobody looked at.

Then read what a photograph says about you once, because the visible frame is only half of what a picture carries — and the other half is already gone from anything you save here.

Do I need permission to post a photograph with somebody else in it?

This is not legal advice and the answer differs by country, by setting and by what the picture is for. What is true everywhere is that the person in the background did not choose to be in your post, and covering them costs you one drag. A school, a workplace and a hospital corridor all raise the stakes further, and in most of those settings somebody has already written a rule about it.

The finder missed somebody.

It will. It is looking for upright, forward-facing, reasonably large faces, and it is set to miss rather than to invent. A face much smaller than about twenty pixels across, turned away, in heavy shadow, or half behind somebody else is the kind it does not see. Every proposal is editable and you can draw the rest by hand in seconds.

Why is the region bigger than the face?

Because a face box is not a head. It stops at the chin and starts at the hairline, which leaves ears, jaw, hair and forehead — plenty to recognise somebody by, particularly if the person looking already knows them. Each proposal is grown outward and lifted before you ever see it.

Is covering a child's face different?

The technique is identical and the tolerance for a mistake is not. For a child, use the solid fill and check every region on the saved copy rather than on the preview. A mosaic over a child's face in a photograph that also names the school and the year group is not really concealment at all.