MOSIP OCR Field Extraction & Verification

This project provides an end-to-end flow for document OCR, field mapping, and verification against an applicant form.

Key Features

• OCR via PaddleOCR for images and PDFs, with dynamic preprocessing and PDF rasterization
• Document-type aware field mapping (Aadhaar/Voter → Name, DL/Passport → Address, Birth/SLC → DOB, Handwritten → All)
• Robust normalization and fuzzy verification for names, addresses, phones, DOB, and gender
• Simple frontend to collect applicant details, upload docs, and review confidence
• Inline edit/save of extracted text per document on the review page; saving re-runs map+verify to refresh confidence scores

• Backend: Python, FastAPI, Uvicorn, PaddleOCR, OpenCV, PyMuPDF (fitz), RapidFuzz, Pydantic
• Frontend: HTML, CSS, Vanilla JS, Bootstrap 5
• Testing: Pytest

• backend/ FastAPI service
  • core/ocr.py OCR utilities: instance cache, preprocess, image/PDF handling
  • core/mapper.py Field mapping: regex + label-value fallback, doc-type filtering
  • core/verifier.py Normalization and scoring + aggregation and decision
  • routes/ FastAPI routes for extraction, mapping, verification
  • app.py FastAPI app with CORS and router setup
• frontend/ Static SPA
  • index.html, styles.css, app.js
• tests/ Pytest suite for mapping/verification logic

• Stub client (backend/core/mosip_client.py) simulates MOSIP pre-registration, upload, and status for offline testing.
• Endpoints (all under /api/v1/mosip):
  • POST /integrate – end-to-end: extract → map/verify (if manual_data provided) → create pre-reg ID → upload document. Default verification threshold is 0.8. Verification uses only overlapping fields between extracted mapped data and manual data; if no overlap, verification is skipped instead of failing.
  • GET /test – quick connectivity stub (returns a canned response).
  • GET /status/{pre_reg_id} – returns stubbed status (pending).
  • POST /batch-submit – process multiple uploaded files; optional verification_data JSON array mirrors manual_data per file.
• Frontend buttons: Submit to MOSIP hits /integrate; Batch Submit posts multiple files to /batch-submit and reports successes/failures.

Use these to validate OCR, mapping, verification, and MOSIP flows.

UI labels (upload panel)

• Name verification document (e.g., Aadhaar, Voter ID)

• Address verification document (e.g., Driving License, Passport)

• Date of birth document (e.g., Birth Certificate, School Leaving Certificate)

• Large file (> size limit) → expect FILE_TOO_LARGE style error if enforced by deployment stack.

• Health: curl http://localhost:8000/api/v1/ocr/health
• Map & verify: curl -X POST http://localhost:8000/api/v1/map-and-verify -H "Content-Type: application/json" -d '{"raw_text":"Name: Alice\nDOB: 2000年01月01日","user":{"name":"Alice","dob":"2000-01-01"}}'
• MOSIP test: curl http://localhost:8000/api/v1/mosip/test
• MOSIP status: curl http://localhost:8000/api/v1/mosip/status/PRE1234567

Backend dependencies (see backend/requirements.txt):

• fastapi, uvicorn
• paddleocr
• numpy, opencv-python
• PyMuPDF (imported as fitz)
• rapidfuzz (optional but recommended)
• unidecode
• pydantic

Create a virtual environment and install dependencies.

python -m venv .venv
.\.venv\Scripts\activate
pip install -r backend\requirements.txt

• Start the backend API (FastAPI + Uvicorn):

cd backend
uvicorn app:app --reload --host 127.0.0.1 --port 8000

• Serve the frontend (static):

cd frontend
python -m http.server 5500

• Open the app at http://127.0.0.1:5500/index.html
• Backend docs: http://127.0.0.1:8000/docs
• MOSIP quick test: GET http://127.0.0.1:8000/api/v1/mosip/test

Base prefix: /api/v1

• OCR
  • POST /api/v1/ocr/extract-text → returns extracted raw text from uploaded file (file)
• Mapping + Verification
  • POST /api/v1/extract-fields → maps fields from raw_text (supports document_type)
  • POST /api/v1/map-and-verify → maps and verifies against user, filtered by document_type
• Direct Verification
  • POST /api/v1/verification/verify → verifies ocr vs user payloads
• MOSIP
  • POST /api/v1/mosip/integrate → single-file OCR → map/verify (optional) → pre-reg stub → upload stub
  • POST /api/v1/mosip/batch-submit → multi-file submit with optional per-file verification data
  • GET /api/v1/mosip/test → connectivity stub
  • GET /api/v1/mosip/status/{pre_reg_id} → status stub

• aadhar or voter: extract and show only name; confidence equals name match vs form
• dl or passport: extract and show only address; confidence equals address match
• birth or slc: extract and show only dob; confidence equals date match
• handwritten: show all fields found; confidence aggregates available fields

• Combines English and Hindi regex patterns for core fields
• Fallback label-value parser for lines like Name: John Smith, Address: 123 Elm St, Phone number: 555-12345, etc.
• Document-type filtering ensures the review displays only relevant fields
• Cleaning:
  • Phone → digits only (strip punctuation/spaces)
  • Email → lowercase, trim spaces, fix common typos (qmail → gmail)
  • Pincode → digits only, up to 6

• Normalization:
  • Unicode (NFKD + transliteration via unidecode)
  • Names/Addresses → lowercase, punctuation removal, abbreviation expansion (e.g., st. → street), stopword removal
  • Phone → digits-only; NSN (last 10 digits) logic
  • DOB → parse common formats to ISO YYYY-MM-DD
• Scoring per field (0..1): name, address, phone, dob, gender
• Aggregation:
  • Weighted average over available fields (default weights: name 0.35, dob 0.30, phone 0.15, address 0.15, gender 0.05)
  • Decision thresholds: MATCH (>=0.85), REVIEW (>=0.6), else MISMATCH

• Applicant form with required fields (all except Middle Name)
• Document uploads per type; processes OCR then mapping+verification
• Review panel shows mapped-only fields per doc type and confidence badge
• Edit/Save UX: only Edit on top; Save appears at bottom when editing applicant data
• Review page extracted-text panel now has Edit/Save to adjust OCR text per doc and recompute confidence

• Run tests:

pytest -q

• If PaddleOCR flags differ across versions, core/ocr.py handles new vs old APIs.
• CORS is open by default; adjust backend/app.py for production.

Proprietary project. Do not redistribute without permission.