Anurag Banerjee
Model added
3a32bd4
Raw
History Blame Contribute Delete
28 kB
"""
FastAPI Certificate Verification API
Seamlessly integrates with any website frontend
"""
from fastapi import FastAPI, File, UploadFile, HTTPException, Header, Query
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
from typing import Optional, List, Dict, Any
import uvicorn
import tempfile
import os
import logging
import time
import asyncio
import re
# Import existing components
try:
from ocr_client import OCRClient
# Use Supabase cloud database instead of local SQLite
from verifier_supabase import SupabaseCertificateVerifier as CertificateVerifier
from yolo_seal_detector import YOLOSealDetector
from vit_seal_classifier import ViTSealClassifier
from image_annotator import annotate_certificate_image, create_annotated_image_url, crop_detected_seals
COMPONENTS_AVAILABLE = True
except ImportError as e:
logging.error(f"Failed to import: {e}")
COMPONENTS_AVAILABLE = False
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
def verify_subject_grades(reg_no: str, ocr_text: str, verifier_instance) -> Dict[str, Any]:
"""
Verify subject grades and SGPA/CGPA from OCR text against database.
Args:
reg_no: Registration number
ocr_text: Extracted OCR text
verifier_instance: CertificateVerifier instance
Returns:
Dictionary with subject verification results
"""
result = {
"subjects_found": False,
"subject_count": 0,
"subjects": [],
"summary": None,
"gpa_verification": {
"sgpa_match": None,
"cgpa_match": None,
"sgpa_db": None,
"cgpa_db": None,
"sgpa_ocr": None,
"cgpa_ocr": None,
"verification_status": "not_checked"
}
}
if not reg_no or not verifier_instance:
return result
try:
# Lookup subjects from database
if hasattr(verifier_instance, '_lookup_subjects'):
subjects = verifier_instance._lookup_subjects(reg_no)
if subjects:
result["subjects_found"] = True
result["subject_count"] = len(subjects)
result["subjects"] = subjects
# Lookup summary (credits, SGPA, CGPA)
if hasattr(verifier_instance, '_lookup_summary'):
summary = verifier_instance._lookup_summary(reg_no)
if summary:
result["summary"] = summary
db_sgpa = summary.get('sgpa')
db_cgpa = summary.get('cgpa')
result["gpa_verification"]["sgpa_db"] = db_sgpa
result["gpa_verification"]["cgpa_db"] = db_cgpa
# Extract SGPA/CGPA from OCR text
# Pattern 1: "SGPA CGPA 9.95 9.78" (both on same line)
combined_match = re.search(r'SGPA\s+CGPA\s+([0-9.]+)\s+([0-9.]+)', ocr_text, re.IGNORECASE)
ocr_sgpa = None
ocr_cgpa = None
if combined_match:
try:
ocr_sgpa = float(combined_match.group(1))
ocr_cgpa = float(combined_match.group(2))
except ValueError:
pass
else:
# Pattern 2: Separate patterns
sgpa_match = re.search(r'SGPA[:\s]+([0-9.]+)', ocr_text, re.IGNORECASE)
cgpa_match = re.search(r'CGPA[:\s]+([0-9.]+)', ocr_text, re.IGNORECASE)
if sgpa_match:
try:
ocr_sgpa = float(sgpa_match.group(1))
except ValueError:
pass
if cgpa_match:
try:
ocr_cgpa = float(cgpa_match.group(1))
except ValueError:
pass
result["gpa_verification"]["sgpa_ocr"] = ocr_sgpa
result["gpa_verification"]["cgpa_ocr"] = ocr_cgpa
# Compare values (tolerance of 0.1)
if ocr_sgpa is not None and db_sgpa is not None:
sgpa_diff = abs(db_sgpa - ocr_sgpa)
result["gpa_verification"]["sgpa_match"] = sgpa_diff < 0.1
result["gpa_verification"]["sgpa_difference"] = round(sgpa_diff, 2)
if ocr_cgpa is not None and db_cgpa is not None:
cgpa_diff = abs(db_cgpa - ocr_cgpa)
result["gpa_verification"]["cgpa_match"] = cgpa_diff < 0.1
result["gpa_verification"]["cgpa_difference"] = round(cgpa_diff, 2)
# Determine overall verification status
sgpa_ok = result["gpa_verification"]["sgpa_match"]
cgpa_ok = result["gpa_verification"]["cgpa_match"]
if sgpa_ok is not None or cgpa_ok is not None:
if sgpa_ok is True and cgpa_ok is True:
result["gpa_verification"]["verification_status"] = "verified"
elif sgpa_ok is False or cgpa_ok is False:
result["gpa_verification"]["verification_status"] = "mismatch_detected"
elif sgpa_ok is True or cgpa_ok is True:
result["gpa_verification"]["verification_status"] = "partial_match"
else:
result["gpa_verification"]["verification_status"] = "unable_to_verify"
except Exception as e:
logger.error(f"Subject verification error: {e}")
result["error"] = str(e)
return result
# Initialize FastAPI
app = FastAPI(
title="Certificate Verification API",
description="AI-Powered Certificate Authentication",
version="1.0.0",
docs_url="/docs",
redoc_url="/redoc"
)
# CORS - Allow any website
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# Global models (loaded once)
yolo_detector = None
vit_classifier = None
ocr_client = None
verifier = None
MODELS_LOADED = False
@app.on_event("startup")
async def startup_event():
"""Load models at startup"""
global yolo_detector, vit_classifier, ocr_client, verifier, MODELS_LOADED
if not COMPONENTS_AVAILABLE:
logger.error("Components unavailable")
return
# Check if we should skip heavy model loading (for free tier with limited RAM)
skip_models = os.getenv("SKIP_MODEL_LOADING", "false").lower() == "true"
try:
logger.info("Initializing API components...")
if skip_models:
logger.warning("Skipping AI model loading (SKIP_MODEL_LOADING=true)")
logger.info("API will run with OCR and database verification only")
yolo_detector = None
vit_classifier = None
else:
# Load YOLO detector from Hugging Face
logger.info("Loading YOLO model from Hugging Face...")
yolo_detector = YOLOSealDetector()
if hasattr(yolo_detector, 'load_model'):
yolo_detector.load_model()
logger.info("YOLOv8 loaded and ready")
# Load ViT classifier from Hugging Face
logger.info("Loading ViT model from Hugging Face...")
vit_classifier = ViTSealClassifier()
if hasattr(vit_classifier, 'load_model'):
vit_classifier.load_model()
logger.info("ViT classifier loaded and ready")
# Initialize OCR client (lightweight)
ocr_client = OCRClient()
logger.info("OCR client initialized")
# Initialize database verifier (lightweight)
verifier = CertificateVerifier()
logger.info("Database verifier initialized")
MODELS_LOADED = True
logger.info("API ready for requests!")
except Exception as e:
logger.error(f"Model loading failed: {e}")
import traceback
traceback.print_exc()
MODELS_LOADED = False
@app.get("/")
async def root():
"""Root endpoint"""
return {
"message": "Certificate Verification API",
"version": "1.0.0",
"status": "online",
"models_loaded": MODELS_LOADED,
"endpoints": {
"verify": "POST /api/verify",
"health": "GET /health",
"docs": "GET /docs"
}
}
@app.get("/health")
async def health_check():
"""Health check"""
return {
"status": "healthy" if MODELS_LOADED else "loading",
"models": {
"yolo": yolo_detector is not None,
"vit": vit_classifier is not None,
"ocr": ocr_client is not None,
"db": verifier is not None
}
}
async def verify_single_certificate(
file_bytes: bytes,
filename: str,
enable_seal_verification: bool = True,
return_annotated_image: bool = False
) -> dict:
"""
Internal function to verify a single certificate
Args:
file_bytes: Certificate image bytes
filename: Original filename
enable_seal_verification: Enable AI seal detection
return_annotated_image: Include annotated image with seal boxes in response
Returns:
dict with verification results
"""
start_time = time.time()
if not MODELS_LOADED:
raise HTTPException(503, "Models loading, try again")
if len(file_bytes) > 10 * 1024 * 1024:
raise HTTPException(400, "File too large (max 10MB)")
if len(file_bytes) == 0:
raise HTTPException(400, "Empty file")
try:
# Create temp file
temp_dir = tempfile.mkdtemp()
file_ext = filename.split('.')[-1] if '.' in filename else 'jpg'
temp_path = os.path.join(temp_dir, f"cert_{int(time.time())}.{file_ext}")
with open(temp_path, 'wb') as f:
f.write(file_bytes)
# Step 1: OCR
logger.info("Running OCR...")
ocr_result = ocr_client.extract_text_from_bytes(file_bytes, language='eng')
if not ocr_result.get('success'):
return JSONResponse(
status_code=200,
content={
"success": False,
"error": "OCR failed",
"message": ocr_result.get('error', 'Text extraction failed')
}
)
# Step 2: Database verification
logger.info("Verifying database...")
verification_result = verifier.verify_certificate(ocr_result, filename)
# Step 3: Seal detection
seal_result = None
seal_detections = []
if enable_seal_verification:
logger.info("Detecting seals...")
try:
summary = yolo_detector.get_detection_summary(temp_path, confidence_threshold=0.5)
seal_detections = summary.get('detections', [])
logger.info(f"YOLO detected {len(seal_detections)} seals")
if summary['total_seals'] > 0:
fake_count = summary['class_distribution'].get('fake', 0)
true_count = summary['class_distribution'].get('true', 0)
avg_confidence = summary['average_confidence']
if fake_count > true_count:
seal_status = "Fake"
status = "Fail"
reason = f"Detected {fake_count} fake vs {true_count} authentic seals"
elif true_count > 0 and fake_count == 0:
seal_status = "Real"
status = "Pass"
reason = f"All {true_count} seals appear authentic"
else:
seal_status = "Suspicious"
status = "Warning"
reason = f"Mixed: {true_count} authentic, {fake_count} fake"
seal_result = {
"status": status,
"seal_status": seal_status,
"reason": reason,
"confidence": avg_confidence,
"total_seals": summary['total_seals'],
"authentic_seals": true_count,
"fake_seals": fake_count,
"detection_method": "YOLOv8",
"individual_predictions": []
}
# Build individual predictions with institution info
for i, detection in enumerate(seal_detections):
det_class = detection.get('class', 'unknown')
det_confidence = detection.get('confidence', 0.0)
if det_class.lower() == 'true' or det_class.lower() == 'real':
pred_status = "Real"
institution = "Visvesvaraya Technological University"
elif det_class.lower() == 'fake':
pred_status = "Fake"
institution = None
else:
pred_status = "Unknown"
institution = None
seal_result["individual_predictions"].append({
"seal_number": i + 1,
"seal_status": pred_status,
"confidence": det_confidence,
"institution": institution,
"bounding_box": detection.get('bbox', detection.get('box', None))
})
else:
seal_result = {
"status": "Warning",
"seal_status": "None Detected",
"reason": "No seals found",
"confidence": 0.0,
"total_seals": 0
}
except Exception as e:
logger.error(f"Seal error: {e}")
seal_result = {"status": "Error", "error": str(e)}
# Final decision
ocr_decision = verification_result.get('decision', 'UNKNOWN')
ocr_confidence = verification_result.get('final_score', 0.0)
# Security first: fake seals = reject
if seal_result and seal_result.get('seal_status') == 'Fake':
final_decision = "FAKE"
confidence = seal_result.get('confidence', 0.0)
reason = "Rejected due to fake seals"
elif ocr_decision == 'AUTHENTIC' and (not seal_result or seal_result.get('status') == 'Pass'):
final_decision = "AUTHENTIC"
confidence = (ocr_confidence + (seal_result.get('confidence', 0) if seal_result else 0)) / 2
reason = "Certificate verified successfully"
elif ocr_decision == 'SUSPICIOUS' or (seal_result and seal_result.get('status') == 'Warning'):
final_decision = "SUSPICIOUS"
confidence = ocr_confidence
reason = "Requires manual review"
else:
final_decision = "FAKE"
confidence = ocr_confidence
reason = "Verification failed"
# Cleanup
try:
os.remove(temp_path)
os.rmdir(temp_dir)
except:
pass
processing_time = time.time() - start_time
# Enrich seal_detections with institution info for annotation
enriched_seal_detections = []
for detection in seal_detections:
det_class = detection.get('class', 'unknown')
enriched_detection = detection.copy()
# Add institution info based on class
if det_class.lower() == 'true' or det_class.lower() == 'real':
enriched_detection['institution'] = "Visvesvaraya Technological University"
else:
enriched_detection['institution'] = None
enriched_seal_detections.append(enriched_detection)
# Generate annotated image and cropped seals if requested
annotated_image = None
cropped_seals = []
if return_annotated_image and enriched_seal_detections:
try:
logger.info("Generating annotated image...")
annotated_image = annotate_certificate_image(file_bytes, enriched_seal_detections)
logger.info("Cropping detected seals...")
cropped_seals = crop_detected_seals(file_bytes, enriched_seal_detections)
except Exception as e:
logger.error(f"Error annotating image: {e}")
# Step 4: Subject grades verification (new feature from main.py)
subject_verification = None
reg_no = verification_result.get('registration_no')
extracted_text = ocr_result.get('extracted_text', '')
if reg_no and verifier:
logger.info("Verifying subject grades...")
subject_verification = verify_subject_grades(reg_no, extracted_text, verifier)
# If GPA mismatch detected with high confidence, flag as suspicious
if subject_verification.get('gpa_verification', {}).get('verification_status') == 'mismatch_detected':
if final_decision == "AUTHENTIC":
final_decision = "SUSPICIOUS"
reason = "GPA values do not match database records - possible tampering"
response_data = {
"success": True,
"decision": final_decision,
"confidence": round(confidence, 3),
"reason": reason,
"processing_time_seconds": round(processing_time, 2),
"details": {
"registration_number": verification_result.get('registration_no'),
"database_match": verification_result.get('db_record') is not None,
"ocr_data": {
"decision": ocr_decision,
"confidence": round(ocr_confidence, 3),
"extracted_text": ocr_result.get('extracted_text', '')[:500],
"field_scores": verification_result.get('field_scores', {})
},
"seal_verification": seal_result,
"subject_verification": subject_verification,
"extracted_fields": verification_result.get('ocr_extracted', {})
},
"filename": filename
}
# Add annotated image to response if generated
if annotated_image:
response_data["annotated_image"] = annotated_image
response_data["annotated_image_url"] = create_annotated_image_url(annotated_image)
# Add cropped seals to response if generated
if cropped_seals:
response_data["cropped_seals"] = cropped_seals
return response_data
except Exception as e:
logger.error(f"Error: {e}")
raise HTTPException(500, f"Verification failed: {str(e)}")
@app.post("/api/verify")
async def verify_certificate(
files: List[UploadFile] = File(...),
enable_seal_verification: bool = Query(True, description="Enable AI seal detection"),
return_image: bool = Query(False, description="Return annotated image with seal bounding boxes")
):
"""
Verify single or multiple certificate images
Args:
files: Certificate image(s) (PNG/JPG/JPEG)
enable_seal_verification: Enable AI seal detection
return_image: Return annotated image with colored boxes around seals (green=authentic, red=fake)
Returns:
JSON with verification results (single format or batch format)
If return_image=true, includes base64 encoded annotated image
"""
if not MODELS_LOADED:
raise HTTPException(503, "Models loading, try again")
# Validate file count
if len(files) > 10:
raise HTTPException(400, "Maximum 10 certificates per request")
# Single file - return original format for backward compatibility
if len(files) == 1:
file = files[0]
if not file.content_type.startswith('image/'):
raise HTTPException(400, f"Invalid file type: {file.content_type}")
file_bytes = await file.read()
try:
result = await verify_single_certificate(
file_bytes,
file.filename,
enable_seal_verification,
return_annotated_image=return_image
)
return result
except Exception as e:
logger.error(f"Single verification error: {e}")
raise HTTPException(500, f"Verification failed: {str(e)}")
# Multiple files - batch processing
batch_start_time = time.time()
results = []
failed_count = 0
# Process files with concurrency limit
semaphore = asyncio.Semaphore(3) # Max 3 concurrent
async def process_one_file(file: UploadFile):
async with semaphore:
try:
# Validate file type
if file.content_type and not file.content_type.startswith('image/'):
return {
"filename": file.filename,
"success": False,
"error": f"Invalid file type: {file.content_type}",
"decision": None
}
# If no content_type, check file extension
if not file.content_type:
ext = file.filename.split('.')[-1].lower() if '.' in file.filename else ''
if ext not in ['jpg', 'jpeg', 'png', 'gif', 'bmp']:
return {
"filename": file.filename,
"success": False,
"error": f"Invalid file extension: {ext}",
"decision": None
}
# Validate file size
file_bytes = await file.read()
if len(file_bytes) > 5 * 1024 * 1024: # 5MB limit per file in batch
return {
"filename": file.filename,
"success": False,
"error": "File too large (max 5MB in batch mode)",
"decision": None
}
if len(file_bytes) == 0:
return {
"filename": file.filename,
"success": False,
"error": "Empty file",
"decision": None
}
# Verify certificate
result = await verify_single_certificate(
file_bytes,
file.filename,
enable_seal_verification,
return_annotated_image=return_image
)
batch_result = {
"filename": file.filename,
"success": True,
"decision": result.get("decision"),
"confidence": result.get("confidence"),
"reason": result.get("reason"),
"processing_time_seconds": result.get("processing_time_seconds"),
"details": result.get("details"),
"error": None
}
# Include annotated image if requested
if return_image and result.get("annotated_image"):
batch_result["annotated_image"] = result.get("annotated_image")
batch_result["annotated_image_url"] = result.get("annotated_image_url")
# Include cropped seals if available
if return_image and result.get("cropped_seals"):
batch_result["cropped_seals"] = result.get("cropped_seals")
return batch_result
except Exception as e:
logger.error(f"Error processing {file.filename}: {e}")
return {
"filename": file.filename,
"success": False,
"error": str(e),
"decision": None
}
# Process all files concurrently
logger.info(f"Processing batch of {len(files)} certificates...")
results = await asyncio.gather(*[process_one_file(f) for f in files])
# Calculate statistics
authentic_count = sum(1 for r in results if r.get("decision") == "AUTHENTIC")
fake_count = sum(1 for r in results if r.get("decision") == "FAKE")
suspicious_count = sum(1 for r in results if r.get("decision") == "SUSPICIOUS")
failed_count = sum(1 for r in results if not r.get("success"))
processed_count = len(results) - failed_count
total_confidence = sum(r.get("confidence", 0) for r in results if r.get("success"))
avg_confidence = total_confidence / processed_count if processed_count > 0 else 0
total_time = time.time() - batch_start_time
return {
"batch": True,
"total_certificates": len(files),
"processed": processed_count,
"failed": failed_count,
"results": results,
"summary": {
"authentic_count": authentic_count,
"fake_count": fake_count,
"suspicious_count": suspicious_count,
"error_count": failed_count,
"total_processing_time_seconds": round(total_time, 2),
"average_confidence": round(avg_confidence, 3)
}
}
@app.post("/api/verify/simple")
async def verify_simple(files: List[UploadFile] = File(...)):
"""Simplified endpoint - just decision"""
result = await verify_certificate(files)
if isinstance(result, dict):
if result.get('batch'):
# Batch response - simplify
return {
"batch": True,
"results": [
{
"filename": r["filename"],
"decision": r.get("decision"),
"confidence": r.get("confidence")
}
for r in result["results"]
]
}
else:
# Single response
return {
"decision": result.get('decision'),
"confidence": result.get('confidence'),
"reason": result.get('reason')
}
return result
@app.get("/api/status")
async def api_status():
"""Detailed status"""
return {
"api_version": "1.0.0",
"models_loaded": MODELS_LOADED,
"components": {
"yolo_detector": {"loaded": yolo_detector is not None, "type": "YOLOv8"},
"vit_classifier": {"loaded": vit_classifier is not None, "type": "ViT"},
"ocr_client": {"loaded": ocr_client is not None, "provider": "OCR.space"},
"database": {"loaded": verifier is not None, "type": "SQLite"}
}
}
if __name__ == "__main__":
port = int(os.getenv("PORT", 8000))
uvicorn.run("api:app", host="0.0.0.0", port=port, reload=False)