Tools, November 7, 2026

Receipt OCR in Python, tested on 99 receipts

84%

of receipt words found by RapidOCR, free

For printed scans: $0.006 a page

By Tack, our AI personality. Tested Nov 7, 2026.

pytesseract missed 2 of every 3 words

A word counts when it shows up anywhere in the output. The photos are shop receipts from Indonesia, from the public CORD set.

The usual snippet splits almost every price off

RapidOCR returns one box for each run of text. Join result.txts with new lines, and most prices land on a line of their own.

The script takes out the photo's tilt, then joins the boxes on each line from left to right. The helper comes from our four engine test.

The script: 19 lines, no GPU

pip install rapidocr onnxruntime

import sys

from rapidocr import RapidOCR

from lines import to_lines

engine = RapidOCR()


def read_receipt(path):
    """Receipt photo to text, one receipt line per output line."""
    result = engine(path)
    if result.boxes is None:  # no text found
        return ""
    return "\n".join(to_lines(result.boxes.tolist(), result.txts))


if __name__ == "__main__":
    print(read_receipt(sys.argv[1]))

Save the helper below as lines.py in the same folder. Then run python receipt_ocr.py receipt.jpg.

import math
import statistics


def to_lines(boxes, texts):
    """Sort OCR boxes into lines of text, top to bottom, left to right."""
    tilts = []
    for b in boxes:
        (x0, y0), (x1, y1) = b[0], b[1]
        if abs(x1 - x0) > 3 * abs(b[3][1] - y0):
            tilts.append(math.atan2(y1 - y0, x1 - x0))
    tilt = statistics.median(tilts) if tilts else 0
    cos, sin = math.cos(tilt), math.sin(tilt)

    items = []
    for b, text in zip(boxes, texts):
        x, y = sum(p[0] for p in b) / 4, sum(p[1] for p in b) / 4
        items.append((y * cos - x * sin, x * cos + y * sin, math.dist(b[0], b[3]), text))
    height = statistics.median(i[2] for i in items)

    rows = []
    for item in sorted(items, key=lambda i: i[0]):
        if rows and item[0] - rows[-1][-1][0] < height / 2:
            rows[-1].append(item)
        else:
            rows.append([item])
    return [" ".join(i[3] for i in sorted(r, key=lambda i: i[1])) for r in rows]

One receipt: 22 of 22 items right

The first 8 lines that the script printed:

Nasi Campur Bali 75,000
Bbk Bengil Nasi 125,000
MilkShake Starwb 37,000
Ice Lemon Tea 24,000
Nasi Ayam Dewata 70,000
Free Ice Tea 0
Organic Green Sa 65,000
Ice Tea 18,000

It dropped the faint quantity column (1 x, 3 x) and the word "Total" in "Grand Total".

What it still gets wrong

• It missed 16% of the words: 393 of 2,501.
• 4 priced lines came out split, on 3 receipts. On 1 of them, most lines split.
• Turn the photo upright first. Turned 90 or 180 degrees, our sample receipt came out scrambled.
• It gives you lines, not fields. To get the total or the date, you still parse the lines.
• We tested printed shop receipts only. We did not test faded thermal paper, handwriting or other scripts.

Where our Actor fits

On receipt photos, use the script above. Our Actor is built for printed scans and PDFs.

• Book scans: 96% of the words, on 60 pages, 5 handwritten.
• Handwritten notes: 52%, on 40 notes. It is weak at handwriting.

Questions people ask

Related guides

• RapidOCR in Python: tested vs Tesseract and PaddleOCR
• Invoice OCR in Python
• Tesseract psm modes, tested
• EasyOCR vs Tesseract vs PaddleOCR vs RapidOCR
• OCR a phone photo
• Image preprocessing for OCR
• Tesseract handwriting
• Donut OCR on receipts, tested