How to Remove a Date Stamp From a Photo, and How the Fill Works

Film cameras from the 80s and 90s, and plenty of early digital ones, burned the date into the corner of every picture: orange digits like 10 30 '26. Scanned family photos are full of them. So are photos from cheap cameras with the setting left on.

Remove it in three steps

1. Open the photo in the image editor and click Remove. (No photo at hand? Pick the sample Photo with a date stamp.)
2. Paint over the digits with the brush. Go a little past the edges: date stamps glow, and the glow needs to go too. Zoom in with 2× or 4× for small ones.
3. Press Remove painted area.

A second later the digits are gone and the water, sky or grass carries on where they were. Not right? Undo, paint a slightly bigger area, and try again.

The same works for dust spots on a scan, a power line across the sky, a small logo you added yourself years ago, or a stranger in the far background.

How the fill works

Blurring the hole from its edges would leave a smooth smudge. The trick is to copy texture instead: fill the hole with small patches, 7 by 7 pixels, taken from elsewhere in the photo, chosen so that they agree with what’s around the hole and with each other. That’s image completion as described by Wexler, Shechtman and Irani, and it runs in three parts.

1. Coarse to fine. The photo is halved in size a few times, until the hole is only a few pixels thick. There, the rough colors and shapes are worked out. Each finer level starts from the coarser one’s answer and adds detail. The number of levels depends on how thick the painted area is, not how wide: a date stamp is wide but its strokes are thin, and halving it too often would merge the digits into one big blob.

2. Finding patches fast: PatchMatch. For every patch that touches the hole, we need the most similar patch from the rest of the photo. Checking them all would take minutes. PatchMatch (Barnes and colleagues, 2009) relies on two facts: neighbors usually match neighbors, so a good match for one patch suggests one for the patch next to it; and random guesses, tried in shrinking circles around the best match so far, find better ones quickly. A few passes of that, and every patch has a good match.

3. Voting, then blending. Each pixel in the hole is covered by many patches, and each says what color it should be. Early on the editor averages those votes. On the last pass of each level it takes the single best patch’s word instead, because averages of slightly different textures turn into flat mush. Finally, copied patches can be a shade lighter or darker than the edge they land next to. The editor measures that mismatch along the edge and spreads a smooth correction across the hole, the way Poisson image editing blends one picture into another. The seam disappears and the copied texture stays.

All of this runs in a Web Worker, so the page stays responsive. On a 12-megapixel photo, a date stamp takes well under a second.

What it doesn’t do

It doesn’t understand the picture. It copies what’s there. If you paint over half a face or the corner of a window, it can only fill with something that looks like nearby texture, not the face or the window. For that you’d want a generative AI model, and that means uploading the photo to whoever runs it.

Key takeaways

• Paint over the date stamp, glow included, and press Remove.
• The fill copies real texture from the photo, coarse to fine, with PatchMatch.
• The edges are blended so the fill matches its surroundings.
• It’s for your own photos: date stamps, spots, wires, your own old marks.

Remove a date stamp →

Need image processing that runs in your users’ browsers, without a server bill per photo? We’ll build it for you, and because we work with AI-assisted development, it costs far less than you might expect. Let’s talk.