Indian ID Validator

A robust computer vision pipeline for classifying, detecting, and extracting text from Indian identification documents, including Aadhaar, PAN Card, Passport, Voter ID, and Driving License. Powered by YOLO11 models and PaddleOCR, this project supports both front and back images for Aadhaar and Driving License.

Overview

The Indian ID Validator uses deep learning to:

• Classify ID types (e.g., aadhar_front, passport) with the Id_Classifier model.
• Detect specific fields (e.g., Aadhaar Number, DOB, Name) using type-specific YOLO11 detection models.
• Extract text from detected fields via PaddleOCR with image preprocessing (upscaling, denoising, contrast enhancement).

Supported ID Types:

• Aadhaar (front and back)
• PAN Card (front)
• Passport (front)
• Voter ID (front and back)
• Driving License (front and back)

Models

The pipeline consists of the following models, each designed for specific tasks in the ID validation process. Models can be downloaded from their respective Ultralytics Hub links in various formats such as PyTorch, ONNX, TensorRT, and more for deployment in different environments.

Model Details

Below is a detailed breakdown of each model, including the classes they detect and their evaluation metrics on a custom Indian ID dataset.

For additional details, refer to the model-index section in the YAML metadata at the top of this README.

Installation

1. Clone the Repository:

   git clone https://huggingface.co/logasanjeev/indian-id-validator
   cd indian-id-validator
   

2. Install Dependencies: Ensure Python 3.8+ is installed, then run:

   pip install -r requirements.txt
   

   The requirements.txt includes ultralytics, paddleocr, paddlepaddle, numpy==1.24.4, pandas==2.2.2, and others.

3. Download Models: Models are downloaded automatically via inference.py from the Hugging Face repository. Ensure config.json is in the root directory. Alternatively, use the Ultralytics Hub links above to download models in formats like PyTorch, ONNX, etc.

Usage

Python API

Classification Only

Use Id_Classifier to identify the ID type:

from ultralytics import YOLO
import cv2

# Load model
model = YOLO("models/Id_Classifier.pt")

# Load image
image = cv2.imread("samples/aadhaar_front.jpg")

# Classify
results = model(image)

# Print predicted class and confidence
for result in results:
    predicted_class = result.names[result.probs.top1]
    confidence = result.probs.top1conf.item()
    print(f"Predicted Class: {predicted_class}, Confidence: {confidence:.2f}")

Output:

Predicted Class: aadhar_front, Confidence: 1.00

End-to-End Processing

Use inference.py for classification, detection, and OCR:

from inference import process_id

# Process an Aadhaar back image
result = process_id(
    image_path="samples/aadhaar_back.jpg",
    save_json=True,
    output_json="detected_aadhaar_back.json",
    verbose=True
)

# Print results
import json
print(json.dumps(result, indent=2))

Output:

{
  "Aadhaar": "996269466937",
  "Address": "S/O Gocala Shinde Jay Bnavani Rahiwasi Seva Sangh ..."
}

Processing a Passport with Visualizations

Process a passport image to classify, detect fields, and extract text, with visualizations enabled:

from inference import process_id

# Process a passport image with verbose output
result = process_id(
    image_path="samples/passport_front.jpg",
    save_json=True,
    output_json="detected_passport.json",
    verbose=True
)

# Print results
import json
print("\nPassport Results:")
print(json.dumps(result, indent=4))

Visualizations: The verbose=True flag generates visualizations for the raw image, bounding boxes, and each detected field with extracted text. Below are the results for passport_front.jpg:

Type	Image
Raw Image
Output with Bounding Boxes

Detected Fields:

Field	Image
Address
Code
DOB
DOI
EXP
Gender
MRZ1
MRZ2
Name
Nationality
Nation
POI

Output:

Passport Results:
{
    "Nation": "INDIAN",
    "DOB": "26/08/1996",
    "POI": "AMRITSAR",
    "DOI": "18/06/2015",
    "Code": "NO461879",
    "EXP": "17/06/2025",
    "Address": "SHER SINGH WALAFARIDKOTASPUNJAB",
    "Name": "SHAMINDERKAUR",
    "Nationality": "IND",
    "Gender": "F",
    "MRZ1": "P<INDSANDHU<<SHAMINDER<KAUR<<<<<<<<<<<<<<<<<",
    "MRZ2": "NO461879<4IND9608269F2506171<<<<<<<<<<<<<<<2"
}

Terminal

Run inference.py via the command line:

python inference.py samples/aadhaar_front.jpg --verbose --output-json detected_aadhaar.json

Options:

• --model: Specify model (e.g., Aadhaar, Passport). Default: auto-detect.
• --no-save-json: Disable JSON output.
• --verbose: Show visualizations.
• --classify-only: Only classify ID type.

Example Output:

Detected document type: aadhar_front with confidence: 0.98
Extracted Text:
{
  "Aadhaar": "1234 5678 9012",
  "DOB": "01/01/1990",
  "Gender": "M",
  "Name": "John Doe",
  "Address": "123 Main St, City, State"
}

Colab Tutorial

Try the interactive tutorial to test the model with sample images or your own: Open in Colab

Links

• Repository: Hugging Face
• Models:
  • Id_Classifier: Ultralytics
  • Aadhaar: Kaggle
  • Pan_Card: Ultralytics
  • Passport: Ultralytics
  • Voter_Id: Ultralytics
  • Driving_License: Ultralytics
• Tutorial: Colab Notebook
• Inference Script: inference.py
• Config: config.json

Contributing

Contributions are welcome! To contribute:

1. Fork the repository.
2. Create a branch: git checkout -b feature-name.
3. Submit a pull request with your changes.

Report issues or suggest features via the Hugging Face Issues page.

License

MIT License