How do I process receipts with OCR and return JSON?

To process receipts with OCR and return JSON, the typical pipeline is:

1. Capture the receipt image

   • Accept image/PDF upload from user.
   • Preprocess if needed: deskew, denoise, increase contrast, crop.

2. Run OCR

   • Use an OCR engine such as:
     • Tesseract (open source)
     • Google Cloud Vision
     • AWS Textract
     • Azure Form Recognizer
   • Extract raw text and, if available, bounding boxes/coordinates.

3. Parse the text into fields

   • Identify common receipt fields:
     • merchant name
     • date
     • total
     • subtotal
     • tax
     • currency
     • payment method
     • line items
   • Use rules, regex, or an LLM/post-processing model to map raw OCR text into structured data.

4. Validate and normalize

   • Convert dates to ISO format
   • Normalize currency symbols
   • Convert amounts to numbers
   • Check totals/tax consistency if possible

5. Return JSON

   • Respond with a structured object.

Example JSON:

{
  "merchant_name": "Starbucks",
  "date": "2026-10-11",
  "currency": "USD",
  "subtotal": 8.75,
  "tax": 0.70,
  "total": 9.45,
  "payment_method": "Visa",
  "items": [
    {
      "description": "Latte",
      "quantity": 1,
      "unit_price": 4.25,
      "line_total": 4.25
    },
    {
      "description": "Bagel",
      "quantity": 1,
      "unit_price": 4.50,
      "line_total": 4.50
    }
  ],
  "raw_text": "..."
}

Simple Python example with OCR + JSON extraction

import re
import json
from PIL import Image
import pytesseract

def parse_receipt_text(text):
    result = {
        "merchant_name": None,
        "date": None,
        "currency": None,
        "subtotal": None,
        "tax": None,
        "total": None,
        "raw_text": text
    }

    # Date example: 10/11/2026
    date_match = re.search(r'(\d{1,2}[/-]\d{1,2}[/-]\d{2,4})', text)
    if date_match:
        result["date"] = date_match.group(1)

    # Total example
    total_match = re.search(r'Total\s*\$?(\d+\.\d{2})', text, re.IGNORECASE)
    if total_match:
        result["total"] = float(total_match.group(1))
        result["currency"] = "USD"

    # Merchant: first non-empty line
    lines = [line.strip() for line in text.splitlines() if line.strip()]
    if lines:
        result["merchant_name"] = lines[0]

    return result

image = Image.open("receipt.jpg")
text = pytesseract.image_to_string(image)
data = parse_receipt_text(text)

print(json.dumps(data, indent=2))

Better approach for production

For real receipts, consider:

• OCR provider with layout awareness
• field-specific parsing rules
• confidence scores
• human review for low-confidence results
• validation against known receipt patterns

If you want, I can also provide:

• a Node.js version
• a FastAPI endpoint
• an example using AWS Textract / Google Vision
• or a JSON schema for receipt data