How to photograph a receipt for better OCR
Most scanning failures are photography failures. Recognition engines expect something close to a flat, evenly lit, high contrast page of text. A phone snap of a receipt curling on a restaurant table is none of those things.
Here is what actually changes the outcome, roughly in order of impact.
Fill the frame
Resolution spent on your table is resolution not spent on the text. Recognition quality depends on how many pixels tall each character is; around 20 pixels is the point below which accuracy drops off a cliff. Getting closer is the single cheapest improvement available to you.
If a receipt is too long to fill the frame in one shot, scan it in two overlapping halves rather than backing away from it.
Flatten it
A curled receipt distorts the baseline of every line of text. Engines segment a page into lines before recognising characters, and a curve makes that segmentation guess wrong. Smooth the paper against the table and hold the edge down if you need to.
This also matters more than it sounds for the totals block, which is usually near the bottom edge where receipts curl most.
Shoot straight down
Photographing at an angle turns a rectangle into a trapezoid and stretches characters unevenly across the frame. Get the camera parallel to the paper. Advanced scanners correct perspective automatically by detecting the document outline, but no correction is as good as not needing one.
Diffuse the light
Two failure modes, opposite causes:
- Glare. A ceiling light reflecting off glossy thermal paper wipes out a band of text completely. There is no recovering it afterwards. Move, or shade the paper with your other hand.
- Shadow. Your own head blocking the light leaves half the receipt darker than the other half. This one is partly fixable, because local thresholding compares each pixel to its neighbours rather than to one global cutoff, but it still costs accuracy.
Indirect daylight is close to ideal. Direct flash on glossy paper is close to the worst case.
Let the camera focus
Motion blur and misfocus both smear character edges, and thin receipt fonts have little margin. Tap to focus, wait for the confirmation, then shoot. A tripod is overkill; bracing your elbows on the table is not.
Know when the receipt is the problem
Some bills will not read well no matter how you photograph them:
- Thermal paper that has faded to grey on grey has lost the contrast the engine needs
- Heavy background graphics or watermarks behind the text confuse line segmentation
- Dot matrix printing with broken character strokes is genuinely hard
- Handwritten amounts are a different problem class entirely, and standard OCR will not read them
For these, scan what you can and correct the rest on the confirmation screen. Faster than fighting the photo.
PDF invoices (no camera needed)
If the receipt already arrived as a PDF — bank statement line, portal invoice, emailed bill — choose Photos or PDF on the scan screen instead of re-photographing the screen. Each page becomes its own receipt to confirm (same batch limit as photos). Password-protected PDFs need unlocking first; everything still stays on your device.
A quick checklist
- Flat on a surface, no curl
- Camera parallel, straight down
- Receipt fills the frame
- No glare band, no hard shadow
- Focus locked before the shutter
- Check the flagged fields afterwards regardless
- Prefer the original PDF when you have one