Every document, every script
Thousands of ID designs across 190+ countries, with complex fonts, diacritics, symbols, and different reading directions. General models never see enough of them to learn.
The problem
General OCR wasn't built for identity.
Off-the-shelf OCR is trained on clean, printed text. Identity documents are the opposite: hundreds of layouts, dozens of scripts, security fonts, and photos taken in the real world.
Real-world capture
Glare, blur, low light, tears, and odd angles wreck accuracy for tools tuned to flatbed scans and clean screenshots.
Fonts, symbols, and the MRZ
Security fonts, diacritics, special symbols, PDF417 barcodes, and machine-readable zones trip up OCR that isn't trained for them.
Any script
complex fonts, diacritics, and reading directions, read by one model
In-house
OCR pipeline with no third-party engine to wait on
How it works
From raw capture to structured data.
Every document runs through Incode's full OCR toolkit: captured, classified, read, decoded, and returned as clean, structured fields.
The output
Clean, structured data. Every field.
Every read returns as structured fields your systems can use, each scored for confidence, with the MRZ and barcode decoded and matched against the print.
- A score on every field Each value ships with a confidence score, so you auto-accept the clean reads and route only the rest to review.
- MRZ and barcode decoded PDF417 and the machine-readable zone are restored from poor captures and cross-checked against the printed fields.
- NFC where it exists The encrypted chip in e-passports and modern IDs is read for the highest data assurance available.
Pass
Session ID #6612
Document validation
Document Classification
Pass
Composite Check Digit
Pass
Barcode PDF417 check
Pass
Tamper check
Pass
Screen ID liveness
Warn
Session Score: Pass
OK / WARN / FAIL on every check
Smart capture
Built for the hardest reads
The cleaner the capture, the better the read. Incode's SDK gets every user to a readable frame on the first try, then extracts what's on it.
-
Real-time guidance
Live feedback fixes framing, glare, and focus before the shot, so the read starts clean.
-
Auto-orient and capture
Detects the ID, straightens even upside-down scans, and captures the instant it's readable.
-
Two-stage extraction
Proprietary OCR reads every field, outperforming general-purpose engines on scripts and symbols.
-
Barcode and MRZ
Decodes PDF417 and the MRZ, then matches both against the printed fields.
-
NFC chip reading
Reads the encrypted chip in e-passports and modern IDs for the highest data assurance.
Accuracy
Proven against general-purpose OCR
Purpose-built beats general-purpose. In head-to-head testing on real IDs, Incode read the fields that off-the-shelf OCR missed.
Incode's OCR is trained on identity documents, not generic text, so it holds up on security fonts, dense address lines, and the machine-readable zone, where general engines drop fields or break the checksum entirely.
- Name 92% vs 77% general-purpose OCR
- Document number 97% vs 91% general-purpose OCR
- Address 85% vs 13% general-purpose OCR
- MRZ 97% vs 62-78% general-purpose OCR
Field-level exact-match accuracy in internal benchmarks, Incode vs general-purpose OCR.
Global coverage
Every ID, in every script
Tell Incode which documents to accept. The fonts, layouts, symbols, and scripts are learned and handled automatically.
- Always current The document library grows continuously as governments issue new and redesigned IDs.
- New formats fast Our in-house labeling team teaches Incode a brand-new document layout in days, not vendor cycles.
- Instant classification AI recognizes each ID on sight and loads the right field template automatically.
4,900+
document types
190+
countries and territories
Canada
United States
Mexico
Guatemala
Colombia
Ecuador
Peru
Brazil
Chile
Argentina
United Kingdom
Spain
France
Germany
Italy
Nigeria
Kenya
South Africa
India
Philippines
Indonesia
Japan
Australia
Verified proof
Global banks, fintechs, and marketplaces verify documents with Incode.
4,900+
identity document types read, across 190+ countries.
8 of 10
top U.S. banks choose Incode
4 of 5
top banks in Latin America run on Incode
97%
MRZ and document-number exact-match accuracy
What is OCR in identity verification?
Optical character recognition (OCR) reads the text on an identity document, name, date of birth, document number, expiry, and the machine-readable zone, and returns it as structured data your systems can use.
How accurate is Incode's OCR?
Incode's purpose-built OCR outperforms open-source and general-purpose alternatives on global IDs, in internal benchmarks reading name fields at 92% accuracy versus 77% for general-purpose OCR, because it is trained on identity documents rather than generic text.
Which documents, countries, and languages does it support?
4,900+ document types across 190+ countries and territories, reading Latin and non-Latin scripts including Cyrillic, Arabic, and Asian characters: passports, national IDs, driver's licenses, and residence permits.
Does Incode read barcodes and the MRZ?
Yes. A dedicated reader decodes PDF417 barcodes and machine-readable zones, restores poor-quality captures with a machine-learning model, and cross-checks them against the printed fields to catch mismatches.
Is the OCR built in-house?
Yes. Incode develops its OCR in-house with no third-party engine in the pipeline, which is why new and redesigned document formats can be supported quickly, and why it can run in a fully air-gapped deployment.
What's next
Extract clean data from any ID.
See how Incode's in-house OCR reads the documents general tools miss.