Face-ID Structure

A facial recognition pipeline with a supporting Flask API for implementing face ID locks in applications.

Pipeline Overview (Colab Notebook)

Objective: Implement face recognition using DeepFace with MTCNN and RetinaFace as detectors.

Dependencies: Requires OpenCV, DeepFace, and associated libraries.

Execution:

• Preprocessing: Prepare video and image inputs for processing (extracting sutable frames from video before converting to embeddings)
• Face Detection: Use MTCNN and RetinaFace for detecting faces.
• Embedding Extraction: Use the ArcFace model to generate face embeddings.

Output: Generates face embeddings and provides a cosine similarity score.

Flask API Conversion

Purpose: Convert the Colab notebook pipeline into a RESTful Flask API for facial recognition applications.

Endpoints:

• /extract_embeddings_from_video: Extracts face embeddings from a video file.
• /extract_embeddings_from_image: Extracts face embeddings from an image file.
• /compare_embeddings: Compares two sets of embeddings and returns a similarity score.

Usage: Accepts video/image inputs, extracts embeddings, and calculates similarity scores.

Usage

1. Run the Flask API:

   python main.py

2. API Endpoints:

   • Extract Embeddings from Video:

     POST /extract_embeddings_from_video

     • Request: Multipart/form-data with 'video' file.
     • Response: JSON with extracted embeddings.

   • Extract Embeddings from Image:

     POST /extract_embeddings_from_image

     • Request: Multipart/form-data with 'image' file.
     • Response: JSON with extracted embedding.

   • Compare Embeddings:

     POST /compare_embeddings

     • Request: JSON with 'embedding1' and 'embedding2'.
     • Response: JSON with similarity score.

   • Home:

     GET /

     • Response: "Welcome message"

Dependencies

• Flask
• OpenCV (Headless)
• DeepFace
• Numpy
• Matplotlib
• Pillow

Future Enhancements

• Real-time video processing.
• Scalability improvements.
• Additional face recognition models.
• Streamline the API to not only provide embeddings but also perform face-ID locking.

Acknowledgments

• DeepFace: A lightweight face recognition and facial attribute analysis framework for Python.