How OCR detects tampered or fake documents

21 Jun 2026·Mahad OCR·3 min read

Reading a document is only half the job. The other half is knowing whether to trust it. Photoshopped passports and edited invoices are a real problem — so Mahad OCR runs a tamper panel on every upload and returns a single tamper score.

The 5 checks

• EXIF inspection: editing software tags (Photoshop, GIMP) in the image metadata.
• MRZ checksum: passport machine-readable zones have built-in check digits that must compute correctly.
• Error-Level Analysis (ELA): re-saved or pasted regions stand out under recompression.
• Background noise: spliced or synthetic areas break the natural sensor-noise pattern.
• Re-upload detection: a previously blocked image is recognised by its hash and rejected.

One score, clear actions

The checks combine into a tamper score from 0 to 1. Clean documents pass; suspicious ones are routed to human review; clearly fake ones are blocked — with the reason recorded in a hash-chained audit log.

curl -X POST https://api.mahadocr.com/v1/documents/upload \
  -H "X-MahadDoc-API-Key: YOUR_KEY" \
  -F "file=@document.jpg"

# poll the result
curl https://api.mahadocr.com/v1/documents/DOC_xxx \
  -H "X-MahadDoc-API-Key: YOUR_KEY"

Because it runs automatically on every document, you catch fraud before it reaches a person — without slowing down legitimate users.