Using computer vision to verify deductible cleaning invoices with Python

So, let me tell you something weird that happened a few months ago…

I got an invoice for a cleaning service. Looked fine at first. Then I noticed the dates were… off. Like, really off. And the price? Way higher than it should’ve been for what they did. That’s when I thought: There’s gotta be a smarter way to check this stuff. Spoiler: there is.

If you’ve ever run into invoices that just don’t add up—or you’re tired of playing detective with smudged PDFs—this post is for you.

Why computer vision?

You’d think most cleaning invoices are just plain text. But no. Half the time they’re scanned sheets, badly lit photos, or screenshots from who-knows-where. That’s where computer vision comes in—because it lets you “see” the document like a human would (but faster and without missing stuff when you're tired).

When I first played around with OpenCV and Tesseract, I was just trying to pull out a date. Now? I’m extracting client names, service codes, totals, and even validating if it’s a legit deductible expense.

Here’s what you actually need

Let me break this down for you—no tech jargon. These are the five things I used (and how they help):

OpenCV

Think of it like the eyes of your app—it detects images, shapes, edges, whatever you need.
Tesseract OCR

This tool “reads” text from images. Like, it literally sees the words in your scanned invoice.
Pandas (yup, the Python one)

This one's great for organizing extracted data like dates and amounts into neat tables.
Regex

Looks nerdy, works wonders. Helps clean up and filter weird invoice formats.
Invoice template matcher

I built a mini-tool that compares layout structure—useful if you work with repeat clients or vendors.

So how do you actually do it?

Here’s the basic flow I use every time:

Step 1: Grab the invoice

Scan it, screenshot it, whatever. Just make sure it’s clear enough.

Step 2: Run OCR

Feed it through Tesseract. It’ll spit out raw text—sometimes messy, but good enough.

Step 3: Use regex to extract what matters

You’re looking for date patterns (\d{2}/\d{2}/\d{4}), dollar signs, service codes, etc. Regex does the heavy lifting.

import pytesseract
import cv2
import re

# Load and preprocess image
image = cv2.imread("invoice.jpg")
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

# OCR
text = pytesseract.image_to_string(gray)

# Extract date and total amount
date_match = re.search(r'\d{2}/\d{2}/\d{4}', text)
amount_match = re.search(r'\$\d+(\.\d{2})?', text)

print("Date found:", date_match.group() if date_match else "No date found")
print("Amount found:", amount_match.group() if amount_match else "No amount found")

Step 4: Compare with your expected data

Did they clean the office on the date you paid for? Was it a monthly Des Plaines Residential Cleaning or a random Sunday sweep? You can tell if you’ve stored legit dates somewhere.

Step 5: Flag and double-check

If anything seems off (like $400 for one room?), flag it. Maybe even call the vendor—or, in some cases, switch to someone more transparent like Cleaning Services Des Plaines il.

Quick case: A dusty office and a shady invoice

There was this one client—small accounting firm, five desks. The place gets cleaned weekly. Nothing fancy. But one day, they got charged for deep carpet cleaning (which, uh, they don’t even have). Turns out the cleaning company copied and pasted the wrong invoice template.

Thanks to a little OCR magic and layout matching, we spotted the duplicate in under five minutes. Normally, you’d spend hours emailing and arguing. But with a bit of automation? Easy fix.

Funny thing is, they’ve now completely overhauled their vendor tracking. And they’re even optimizing their Office Organization in Des Plaines routines. Win-win.

Tools that helped me (in real life)

I’m not here to pitch products, but here are a few random things that saved my sanity:

Notion + OCR scripts I store scripts and outputs in a Notion page. Easy to search later.
PythonAnywhere Great for testing things on the go.
Airtable Helps track which invoices have been verified.

Why bother doing all this?

Because:

You’ll save time—like, hours per month.
You’ll catch errors before they become tax headaches.
You’ll look pro—clients and coworkers love seeing smart solutions.
And honestly? It just feels cool to solve stuff like a detective with Python.

Try it out—just once

Start with one invoice. Pick a clear photo or scan. Run it through OCR. See what you find.

Maybe you’ll spot a small mistake. Maybe you’ll find a huge billing issue. Either way, it’s a great way to feel in control.

If you’re like me, you’ll wonder why you didn’t do this sooner. So go on—give it a try this week, you’ll see. 😉

DEV Community