Face Verification Lambda Function

This repository contains a Face Verification Lambda Function designed to:

1. Detect human faces in images.
2. AI generated image prevention and detection.
3. Verify if two images belong to the same individual.

The function leverages:

• AuraFace-v1 for face embedding and similarity verification.
• Trained AI Image Classifier (CatBoost) for AI generated Image classification and prevention.

Special thanks to the developers of these models!

---

Features

• Detects faces in provided images.
• Validates that both the album and selfie images contain a single face.
• Validate that the given image is not AI generated Image
• Verifies whether the faces in the two images belong to the same individual.
• Supports both AWS S3 integration and local file paths.

---

Important Note

• Run python model_download.py script to download AuraFace and CatBoost.

---

Requirements

• Docker for local testing.
• Postman or any API testing tool for testing the function.

---

Running Locally

Step 1: Start the Lambda Function in Docker

Run the following command to spin up the Lambda function in a Docker container:

sudo docker run -it --rm --platform linux/amd64 \
  -v ~/.aws-lambda-rie:/aws-lambda \
  -p 9000:8080 \
  --entrypoint /aws-lambda/aws-lambda-rie <your-image-name> \
  /usr/local/bin/python3.11 -m awslambdaric handler.<your-handler-function-name>

Step 2: Test with Postman

1. Open Postman.
2. Use the following URL:

   http://localhost:9000/2015-03-31/functions/function/invocations
   

3. Send a POST request with a JSON body similar to:

{
    "album": "<path-to-album-image>",
    "selfie": "<path-to-selfie-image>"
}

---

Configuration

.env File

Set the following variables in a .env file:

USE_S3=ON                # Set to 'OFF' to use local paths
BUCKET_NAME=<bucket-name>
KEY_ID=<aws-key-id>
SECRET=<aws-secret>
THRESHOLD = 0.5          # WOrks fine on 0.5

Using Local Paths

To test locally:

• Set USE_S3=OFF in the .env file.
• Provide absolute paths to the album and selfie images in your request payload.

Switching to AWS S3

For production or Lambda deployment:

• Set USE_S3=ON in the .env file.
• Provide S3 paths for the album and selfie in your request payload.

---

Workflow

1. Input: The function accepts two images:

   • Album: A reference image of the person.
   • Selfie: A selfie image of the same person.

2. Process:

   • Downloads the images from S3 (if USE_S3=ON).
   • Validates that each image contains exactly one face.
   • Validate that given image is not AI generated image.
   • Extracts embeddings and calculates cosine similarity.

3. Output:

   • Success: Confirms if both images are of the same individual.
   • Error: Provides clear error messages if validation fails.

---

Models Used

1. Face Detection & Verification

• AuraFace-v1
  Detect Faces and Generates embeddings and compares cosine similarity to verify identity.
• In-House Trained AI Image Classifier (CatBoost)
  AI generated Image classification and prevention.

---

Troubleshooting

• Images Not Found: Ensure the provided file paths or S3 keys are correct.
• Multiple Faces Detected: Ensure only one person is visible in each image.
• AI Image detected: Do not use AI generated image. AI generated Image is not acceptable.
• Similarity Check Fails: Images of the same individual may differ due to lighting or angles; try clearer images.

---

Future Work

• Develop methods to prevent the face validation pipeline from accepting faces displayed on a screen.

---

License & Acknowledgments

• Face Verification Model: AuraFace-v1
• AI Image Classification (In-house): Trained AI Image Classifier (CatBoost)

Thank you to the contributors of these models for making them publicly available.