Singapore Receipt OCR: Formats, Items and Totals

Understand the printed transaction first

Keep printed GST amounts distinct from bases and totals, and distinguish payment receipts from stored-value balances. Start by checking whether the document is a completed payment, an estimate, an order confirmation or an account summary. A successful read of the header is not enough: the amount beside the final payable label and the boundaries of individual item rows are the evidence that makes a useful expense record.

Extract receipt fields without inventing a breakdown

Use the receipt OCR API for merchant_name, date, time, currency, total_amount, tax_amount and items. Each item exposes name, quantity and price; price is the whole-row amount, not necessarily a unit price. The public schema has no dedicated tender, tip, loyalty, QR payload or receipt-number object. Keep the source document when those details matter. Related format guides: FairPrice receipt OCR.

From receipt image to an expense record

• Check the document-specific layout: Keep printed GST amounts distinct from bases and totals, and distinguish payment receipts from stored-value balances.
• Save the purchase once: Associate the image with an application-generated receipt ID. A payment slip and an itemized receipt can describe the same purchase; merchant, date and amount can support a duplicate review but are not a universally unique key.
• Export item rows to CSV or Excel: Create one row per returned item and repeat the parent receipt ID, merchant and date. Sum price only where it represents a resolved row amount; do not multiply it by quantity again.
• Route uncertain records for review: Inspect success, is_valid, status, issues and confidence before accepting the extracted data. Keep unresolved quantities or totals unresolved. An HTTP success code does not establish that every amount is readable.

Read a receipt from a Python backend

Submit one receipt image using the standard endpoint. Keep the API key server-side and examine quality indicators before mapping the result to an expense. The sample does not claim a live test on this merchant.

Prerequisite: Python 3, pip install requests; STRUCTOCR_API_KEY environment variable

import base64
import os
import requests

with open("receipt.jpg", "rb") as source:
    encoded = base64.b64encode(source.read()).decode("ascii")

response = requests.post(
    "https://api.structocr.com/v1/receipt",
    headers={"x-api-key": os.environ["STRUCTOCR_API_KEY"]},
    json={"img": encoded},
    timeout=60,
)
response.raise_for_status()
result = response.json()
if not result.get("success"):
    raise RuntimeError("Extraction was not successful; inspect the response securely")
data = result.get("data") or {}
needs_review = (
    not data.get("is_valid")
    or data.get("status") != "complete"
    or bool(data.get("issues"))
    or data.get("confidence") == "low"
)
# Persist through your own storage layer; avoid logging receipt contents.
print({"needs_review": needs_review, "item_count": len(data.get("items") or [])})

Synthetic receipt JSON example

Illustrative values only, not output from a local sample or a merchant benchmark. Item price is a line amount. Currency must be resolved from the submitted receipt rather than hard-coded from this example.

{
  "success": true,
  "data": {
    "response_version": 2,
    "is_valid": true,
    "merchant_name": "Example Merchant",
    "date": "2026-09-01",
    "time": "12:30",
    "currency": "SGD",
    "total_amount": 12,
    "tax_amount": 1,
    "items": [
      {
        "name": "EXAMPLE PURCHASE",
        "quantity": 2,
        "price": "8.00"
      },
      {
        "name": "EXAMPLE CHARGE",
        "quantity": 1,
        "price": "3.00"
      }
    ],
    "confidence": "high",
    "validation_error": null,
    "status": "complete",
    "issues": []
  }
}