What a scanner does that a camera does not
A flatbed scanner presses the page flat, lights it evenly, and photographs it from directly above. A phone held at arm's length does none of those things. The page comes out as a trapezoid rather than a rectangle, one side is brighter than the other, and there is usually a shadow where your hand was.
This tool corrects all three. It finds the four corners of the page, applies the transform that maps them back onto a true rectangle, and evens out the lighting. What comes out looks like a scan.
Why the correction needs a perspective transform
It is tempting to think rotating and cropping would be enough. It is not. When you photograph a page at an angle, the edge further from the camera appears shorter than the near edge — the rectangle becomes a trapezoid, and the text tapers along with it. Rotation cannot undo that, because rotation preserves parallel lines and perspective is precisely what breaks them.
What is needed is a homography: an eight-parameter transform computed from four point correspondences. Every pixel in the output is traced back through that transform to find where it came from in the photograph, then sampled with bilinear interpolation so small text stays smooth instead of turning into jagged edges.
Why the corner handles are always there
Automatic detection thresholds the image and looks for the largest bright region, which works nicely when the page sits on a dark desk and poorly when it sits on a white table. Rather than pretend otherwise, the four handles are always draggable, and the tool tells you when detection failed instead of silently using a bad guess.
Choosing a clean-up mode
| Mode | Best for | File size |
|---|---|---|
| Scan | Printed text, forms, letters | Smallest |
| Greyscale | Pencil, faint print, photographs of text | Medium |
| Colour | Coloured forms, stamps, signatures, ID cards | Largest |
Scan mode does more than raise the contrast. It estimates the background brightness across the page and divides the image by it, which cancels uneven lighting before the contrast is applied. A single global threshold cannot do that — one shadow across the corner and half the text disappears with it.
Getting a good photograph
- Put the page on a surface that contrasts with it. A white page on a dark desk detects almost every time; white on white rarely does.
- Photograph from directly above, with the whole page and a little margin in frame.
- Avoid casting your own shadow across the page — light from the side rather than from behind you.
- Flatten folds first. The transform assumes the page is flat; it cannot undo a curl.
- Hold steady. Blur is the one thing no amount of correction recovers.
Multiple pages
Scan a page, press Add page, then load the next photograph. Pages collect in order and export together as a single PDF, each sized to its own aspect so nothing is stretched. Remove any page that did not come out well before exporting.
Nothing is uploaded
Corner detection, the perspective transform, clean-up and the PDF itself are all produced in your browser. The document never leaves your device.
That is the substantive difference from a phone scanning app. The popular ones want an account, camera-roll access and often a subscription, and more than one has been found sending documents to its own servers. The things people scan — contracts, bank statements, certificates, passports — are exactly the things you would least want on someone else's machine.