OCR explained: what it is, which tools do it, and how to use it on receipts and invoices
Short answer
OCR, optical character recognition, turns the text in a picture or scanned PDF into text a computer can search and copy. Free options include Tesseract, which is open source, and Google Cloud Vision, which lists its first 1,000 units a month as free. Cloud services such as Amazon Textract also read receipts and invoices into fields. A person should check every extracted amount before it reaches a spreadsheet.
On this page
OCR facts checked
- What it does
- Converts text in images and scanned PDFs into machine-readable text
- Free options read
- Tesseract (open source, Apache License 2.0). Google Cloud Vision, first 1,000 units a month free
- Cloud price examples
- Google Vision text detection $1.50 per 1,000 units after the free tier. Amazon Textract Detect Document Text $1.50 per 1,000 pages. Prices in US dollars
- Not quoted
- Azure Document Intelligence rates (the page showed placeholders). Adobe Acrobat plan prices (not read)
- Exchange rate
- USD 1 = PHP 62.821, as of 9 Oct 2026
- Last verified
- 12 Oct 2026
How OCR works, in plain words
An OCR program takes an image and does three jobs. It finds where the text sits on the page. It works out which character each shape is. Then it returns the characters as text, often with the position of each word. Tesseract, the open-source engine, describes a neural-network engine for recognising lines of text. Google’s Vision documentation says its document mode returns results structured by page, block, paragraph and word.
Plain OCR returns text and nothing else. A second kind of tool, usually called document AI or intelligent document processing, goes further. It reads a receipt or invoice and returns named fields such as vendor, date and total. Microsoft describes Document Intelligence as a machine-learning-based OCR and intelligent document processing service that extracts key data from forms and documents. Google describes Document AI as converting unstructured document data into structured fields such as database-ready values.
Quality depends on the image. A flat, well-lit photo reads better than a crumpled, shadowed one. Language settings matter too, and handwriting is harder than print.
How to OCR a PDF
A scanned PDF often has no selectable text. To make it searchable you run OCR, which adds a text layer behind the page image so the page looks the same.
- Adobe Acrobat. Adobe’s help page is titled “recognize text” and blocked this site’s reader. Search-result summaries of it describe choosing the page range and document language, then running Recognize Text, with a “Correct recognized text” step for words the software marks as uncertain. Menu names change between versions, so follow Adobe’s current page. Acrobat’s plan prices were not read.
- Tesseract. Free. It runs from the command line on an image and writes out text. A PDF is usually converted to images first.
- Cloud services. Google, Amazon and Microsoft accept images and PDFs through an API. Amazon states that its service analyzes image and PDF files.
If a scan is going to a client or a court, keep the original and save the OCR copy as a second file.
Tools compared, with sourced facts
| Tool | What it does, as documented | Price as listed |
|---|---|---|
| Tesseract | Open-source OCR engine with a command-line program and C and C++ API. Recognises more than 100 languages out of the box | Free (Apache License 2.0) |
| Google Cloud Vision | Text detection and document text detection. Document mode is for dense text and handles handwriting | First 1,000 units a month free, then $1.50 per 1,000 units, about PHP 94 |
| Google Document AI | Digitize (OCR), extract (form parser, pretrained processors) and classify | Not read |
| Amazon Textract | Detects printed and handwritten text. AnalyzeExpense for invoices and receipts. AnalyzeID for U.S. IDs | Detect Document Text $1.50 per 1,000 pages (US West, Oregon). Analyze Expense $10 per 1,000 pages, about PHP 628 |
| Azure Document Intelligence | OCR plus prebuilt invoice, receipt and ID models | Not quoted. Page showed placeholders. Free tier of 0 to 500 pages a month listed |
| Adobe Acrobat | Recognise text in scanned PDFs | Not read |
Swipe the table sideways to see all columns.
Last verified: 12 Oct 2026. Google’s Vision prices are in US dollars. Amazon’s per-1,000 figures are the per-page rates on its page multiplied by 1,000, and its free tier is for new AWS customers for three months, covering 1,000 pages a month of text detection and 100 of expense analysis. Other regions can price differently.
What this means for Philippine receipts and invoices
Typical uses for a small Philippine business:
- Receipts and supplier invoices. Pull date, supplier and total into an accounting sheet.
- Delivery receipts and forms. Turn paper forms from branches or agents into rows.
- ID checks for onboarding. Read the name and number from an ID image. Amazon’s documentation says its ID analysis covers drivers licenses and passports issued by U.S. government, so a Philippine ID is not covered there. For Google and Microsoft the pages read did not name Philippine IDs, so test with your own samples before relying on any of them.
- Searchable archives. Make old scanned contracts and records searchable.
Receipts printed in Taglish or with faded thermal ink may read poorly. No tool was tested for this page, so use a sample of your own documents.
Data Privacy Act points
Receipts, invoices and especially ID images contain personal information. Republic Act No. 10173 sets duties that apply when you send those images to a cloud OCR service:
- Processing must follow transparency, legitimate purpose and proportionality (Section 11).
- Section 13 covers sensitive personal information. Section 3(l) lists information issued by government agencies peculiar to an individual, such as social security numbers, so ID images deserve extra care.
- The controller must use reasonable and appropriate security measures (Section 20).
- A controller remains accountable for data handed to another party for processing (Section 21).
A local tool such as Tesseract keeps the images on your own machine. A cloud service sends them to the vendor, so read its data terms first. This is not legal advice, so ask your data protection officer or a lawyer how the Act applies to you.
An OCR-to-spreadsheet workflow
This outline suits receipts and invoices arriving by email or a shared folder. It uses a cloud document service for fields, and the same shape works with other tools. See the automation pillar guide for how a workflow is built.
Collect the files
Send receipts to one email address or one shared folder, as photos or PDFs. A new file becomes the trigger for the workflow.
Check the image
Reject files that are blurred or cropped. Ask the sender for a flat, well-lit photo.
Run the OCR step
Send the file to an OCR or document service, such as a Textract expense call or a Document AI processor, using an HTTP request or a ready-made node in a tool like n8n or Zapier.
Map the fields
Take vendor, date, tax and total from the result and write them to named columns in a sheet. Keep the original file link in the same row.
Add a review column
Mark each row Unchecked. Where the service gives a confidence value, flag low-confidence rows.
Review the numbers
A person compares each total, date and tax figure against the image before the row counts. Fix errors by hand.
Move reviewed rows on
Only rows marked Checked go to the accounting file or the finance report.
The service replaces typing, and a person still verifies the figures. Similar-looking digits and faded print can be misread, and a wrong total in a ledger carries into every report built on it. Adobe’s page, as summarised in search results, describes a review step for words the software marks as uncertain, which shows that vendors expect some errors.
n8n and Zapier are the workflow tools covered on this site. Neither’s pages were read here for an OCR-specific node, so check their app lists. OCR is different from robotic process automation, which operates software screens. The RPA comparison explains the difference.
Free and online OCR tools
Free choices that were verified: Tesseract on your own computer, and Google Cloud Vision’s first 1,000 units a month. Many websites also offer online OCR for a single file. None was read for this page. Before using one, check what it says about storing your upload, and do not upload IDs or financial records to a site whose terms you have not read.
Frequently asked questions
What is OCR?
Optical character recognition. Software that reads text in an image or scanned PDF and returns it as text you can search and copy.
How do I OCR a PDF?
Open it in a tool that has a text recognition feature, such as Adobe Acrobat, choose the language and page range, and run it. Free options include Tesseract on the command line. Keep the original file.
Is there a free OCR tool?
Tesseract is open source under the Apache License 2.0. Google Cloud Vision lists the first 1,000 units a month as free. Both need some technical setup.
Can OCR read receipts and invoices into a spreadsheet?
Amazon Textract and Azure Document Intelligence document invoice and receipt processing, and Google Document AI lists pretrained processors. Send the output to a sheet through a workflow tool and check the amounts yourself.
Can OCR read Philippine IDs?
Amazon’s ID analysis is documented for U.S. government IDs. The Google and Microsoft pages read did not name Philippine IDs. Test with your own samples.
Is OCR safe for documents with personal data?
It depends on where the processing happens. A cloud service sends the file to the vendor, so apply the Data Privacy Act duties and read the vendor’s data terms.
Sources
All sources were read on 12 Oct 2026, except Adobe’s help page, which blocked the reader and is described from search results.
- Tesseract OCR, GitHub repository: engine, license, languages.
- Google Cloud, Vision pricing and OCR documentation.
- Google Cloud, Document AI overview.
- Amazon Web Services, Textract pricing and What is Amazon Textract?.
- Microsoft Learn, Azure Document Intelligence overview and pricing.
- Adobe, Recognize text: not read directly.
- Republic Act No. 10173, Data Privacy Act of 2012, via Lawphil.
- European Central Bank euro reference rates, retrieved through the Frankfurter API (api.frankfurter.dev): EUR to PHP and USD, 9 Oct 2026.