Receipt OCR: receipts are harder than invoices
Invoices are printed on flat paper at a consistent size. Receipts are thermal-printed, curled, faded and photographed on a table in bad light. Everything that makes OCR hard, receipts have.
Thermal paper fades
A supermarket receipt from six months ago may already be grey. There is nothing to recover — the ink has physically changed. Photograph receipts the day you get them, and you will never fight this.
Curled paper defeats scanners
A curled receipt creates shadows and skew that no preprocessing fully fixes. Flatten it under a book for a minute, or use a phone app with edge detection. The thirty seconds you spend flattening saves minutes of correction.
What to extract from a receipt
For expenses, four fields are enough: date, merchant, total, payment method. Line items are usually not needed for accounting, and trying to capture them multiplies the failure surface. Decide what you actually need before choosing a tool.
Totals hide at the bottom, next to other numbers
A receipt often shows subtotal, discount, tax and total in a cluster. The total is usually the largest number, but a cash-tendered amount can exceed it. If the tool asks you to confirm the total rather than asserting it, that is a design choice worth trusting.
Common questions
Can OCR read a crumpled receipt?
Sometimes. Flattening and even lighting matter more than which engine you use.
Should I keep the paper receipts?
Yes — most tax authorities expect the original records for several years. The scan is a convenience, not a replacement.
Do I need line items?
Rarely for accounting. Category totals are usually enough, and they are far more reliable.