How to scan a damaged QR code

What error correction can and cannot fix

Every QR code carries Reed–Solomon error correction, which can rebuild missing or misread data up to a level chosen when the code was made:

Level	Recoverable data
L	~7%
M	~15% (the common default)
Q	~25%
H	~30%

So a scratch across a level-M code eating 10% of the modules is routinely survivable. But error correction protects the data region only. The three large finder patterns in the corners are how the scanner locates the code at all: destroy one and decoding never starts, regardless of how pristine the rest is. The quiet zone matters the same way: damage that clutters the margin can stop the code being found. The full mechanics are in error correction levels explained.

The rescue sequence

Photograph the damaged code as square-on and evenly lit as you can, then work on the image:

1. Crop to the code plus a small margin.
2. Flatten: use a perspective/skew tool to make the grid square. Crumpled paper responds surprisingly well to this step.
3. Boost contrast aggressively; convert to greyscale and push toward black-and-white. Fading is often just a contrast problem in disguise.
4. Decode after every step with an image-based scanner rather than a live camera, uploads let you iterate on the same frame.

The white-patch trick

Counter-intuitive but effective: cover damaged areas with clean white: paint them out in an editor, or physically stick plain white paper over the ruined region before photographing.

The reasoning has two layers. Practically, a smudge or torn edge produces ambiguous grey mess that skews the decoder's thresholding and can corrupt the reading of undamaged neighbouring modules; a clean white patch confines the damage to the modules actually lost. Theoretically, Reed–Solomon can correct twice as many erasures (unreadable symbols at known positions) as errors (symbols read wrongly), and decoders that flag suspect regions get closer to that erasure budget when the damage is crisp and obvious rather than plausible-looking noise. Never try to redraw missing modules by guesswork: a wrongly guessed module is an error, which costs double what an honest blank does.

When it is genuinely over

• More than ~30% of the code is gone, beyond even level H's budget.
• A finder pattern corner is destroyed and cannot be reconstructed by copying one of the surviving squares (a last-resort editor trick that sometimes works, since all three are identical 7×7 patterns).
• The code was low error correction and dense, long-URL codes at level L have almost no margin.

At that point the only recovery is context: the surviving text of a partially decoded URL, or asking whoever printed it.

Prevention costs less than rescue

Codes destined for weather, handling or abrasion should be generated at level H, printed larger than the minimum, and produced on materials that fail gracefully: see outdoor durability and the print checklist. A code you verify decodes before printing and print with margin survives most of what the world does to it.

FAQ

Can a ripped QR code still be scanned?

Often, yes. Error correction rebuilds up to 7–30% of the data depending on the code's level, provided the three corner finder patterns survive. Photograph it flat, boost the contrast, and cover torn areas with clean white before decoding.

How much of a QR code can be missing?

Up to roughly 30% at error correction level H, 15% at the common default M. Damage to the corner finder patterns or the surrounding quiet zone is not covered and can make even lightly damaged codes unreadable.

Why cover QR code damage with white instead of redrawing it?

A wrong guess counts as an error, and Reed–Solomon corrects twice as many known blanks (erasures) as wrong values. Clean white also stops smudged regions from skewing the threshold used to read undamaged modules.

Can a faded QR code be recovered?

Usually. Fading is mostly a contrast problem: photograph the code evenly lit, convert to greyscale, and push the contrast hard toward black-and-white, then decode the image. If thermal-paper print has faded to nothing, the data is gone.