# 🚀 Face Recognition System Using FaceNet

A high-performance **Face Recognition system** inspired by the FaceNet architecture.  
This project improves accuracy while drastically reducing training data requirements.

---

## 📌 Project Summary

This project implements a deep metric-learning-based face recognition model using a modified version of the traditional triplet loss approach.

### 🎯 Objectives

- Improve embedding discrimination
- Reduce required dataset size
- Enhance model generalization
- Maintain real-world deployment capability

---

## 🧠 Model Architecture

The system learns a compact embedding representation for each face image.  
Faces belonging to the same identity are mapped closer together in embedding space, while different identities are pushed further apart.

---

## 🔬 Custom Loss Function

A modified loss function was introduced to improve separation efficiency:

```math
∑_{i=1}^{N} [ max(|f(x^A) - f(x^P)| - 0.2α) 
+ max(-|f(x^A) - f(x^N)| + α), 0 ]
```

### Where:

- `x^A` → Anchor image  
- `x^P` → Positive image (same identity)  
- `x^N` → Negative image (different identity)  
- `f(x)` → Learned embedding function  
- `α` → Margin parameter  

### 🔎 Key Improvements

- Stronger intra-class compactness  
- Larger inter-class separation  
- Better performance with fewer samples  

---

## 📊 Results

| Metric | Value |
|--------|--------|
| Accuracy Improvement | **+10%** over original benchmark |
| Training Images Used | **6,000** |
| Images Used in Original Paper | **1,000,000** |
| Generalization | Strong performance on unseen identities |

The model achieved higher accuracy while using less than 1% of the dataset size reported in the original paper.

---

## 🌍 Real-World Applications

- 🔐 Access Control Systems  
- 🏢 Smart Attendance Systems  
- 🪪 Identity Verification  
- 🛡 Security Authentication  
- 📱 Edge AI Deployment  

This project demonstrates that well-designed loss engineering and optimization strategies can outperform large-scale data approaches.

---

## 🛠️ Tech Stack

- Python
- PyTorch / TensorFlow (depending on your implementation)
- OpenCV
- NumPy


---


---

## 🔗 GitHub Repository

👉 **View the full project here:**  
https://github.com/mohammedaboallayl/Face_Recognition_From_Scratch

---

## 📬 Contact

If you have questions or collaboration ideas, feel free to connect.