https://github.com/balakrishna-kini/deep-fake-video-detection

DEEP-FAKE VIDEO DETECTION: A DEEP LEARNING APPROACH FOR FACE-SWAP DETECTION
https://github.com/balakrishna-kini/deep-fake-video-detection

deeplearning flask haarcascade-frontalface lstm resnet-50

Last synced: 7 months ago
JSON representation

DEEP-FAKE VIDEO DETECTION: A DEEP LEARNING APPROACH FOR FACE-SWAP DETECTION

• Host: GitHub
• URL: https://github.com/balakrishna-kini/deep-fake-video-detection
• Owner: Balakrishna-kini
• License: mit
• Created: 2025-03-04T15:14:09.000Z (8 months ago)
• Default Branch: master
• Last Pushed: 2025-03-04T18:17:36.000Z (8 months ago)
• Last Synced: 2025-03-04T18:25:50.249Z (8 months ago)
• Topics: deeplearning, flask, haarcascade-frontalface, lstm, resnet-50
• Language: Jupyter Notebook
• Homepage:
• Size: 8.54 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
A deep learning-based web application that detects manipulated (deepfake) videos using a ResNet50-LSTM architecture. This tool helps identify AI-generated or manipulated video content with high accuracy.

##

![frontend](Frontend.png)

## 🌟 Features

- **Advanced Deep Learning Model**: Combines ResNet50 features with LSTM temporal analysis for accurate deepfake detection

- **User-Friendly Interface**: Simple drag-and-drop or file selection for video uploads

- **Real-Time Analysis**: Processes videos and provides immediate results

- **Confidence Metrics**: Shows detection confidence percentage for transparency

- **Responsive Design**: Works on desktop and mobile devices

## 🧠 Model Architecture

The deepfake detection model uses a sophisticated architecture:

- **Feature Extraction**: ResNet50 processes video frames to extract facial features

- **Face Detection**: haarcascade_frontalface for accurate face detection in video frames

- **Temporal Analysis**: LSTM layers analyze sequences of frames to detect inconsistencies over time

- **Classification**: Determines if a video is real or fake with a confidence score

## 🛠️ Technologies Used

### Backend

- Python 3.8+

- TensorFlow/Keras

- Flask

- OpenCV

- NumPy

- haarcascade_frontalface

### Frontend

- HTML5

- CSS3

- JavaScript

- Font Awesome

## 📋 Prerequisites

- Python 3.8 or higher

- pip (Python package manager)

- 8GB+ RAM for video processing

- Modern web browser

- [Dataset](https://www.kaggle.com/datasets/ameencaslam/three-body-dataset)

## 🔧 Installation

1. **Clone the repository**

```shellscript

git clone https://github.com/yourusername/deepfake-detector.git

cd deepfake-detector

```

2. **Create a virtual environment (recommended)**

```shellscript

python -m venv venv

# On Windows

venv\Scripts\activate

# On macOS/Linux

source venv/bin/activate

```

3. **Install dependencies**

```shellscript

pip install -r requirements.txt

```

4. **Download the pre-trained model**

Place the [`best_model.keras`](DeepFake_Video_Detection.ipynb) file in the root directory of the project.

Note: I trained the model using [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/weiji14/deepbedmap/)

## 🚀 Usage

1. **Start the Flask server**

```shellscript

python app.py

```

2. **Access the web interface**

Open your browser and navigate to: `http://127.0.0.1:5000`

3. **Upload a video**

Drag and drop a video file or click to select one from your device.

4. **Analyze the video**

Click "Detect Deepfake" and wait for the analysis to complete.

5. **View results**

The application will display whether the video is real or fake, along with a confidence percentage.

## 📁 Project Structure

```plaintext

deepfake-detector/

├── app.py                  # Flask application

├── deepfake_detector.py    # Core detection functionality

├── requirements.txt        # Python dependencies

├── static/                 # Static files

│   ├── style.css           # CSS styles

│   └── script.js           # JavaScript functionality

├── templates/              # HTML templates

│   └── index.html          # Main page

└── uploads/                # Temporary storage for uploaded videos

```

## 🔍 How It Works

1. **Video Upload**: User uploads a video through the web interface

2. **Frame Extraction**: The system extracts frames from the video

3. **Face Detection**: haarcascade_frontalface detects faces in each frame

4. **Feature Extraction**: ResNet50 extracts features from the face regions

5. **Sequence Analysis**: LSTM layers analyze the temporal consistency

6. **Classification**: The model determines if the video is real or fake

7. **Result Display**: Results are displayed with confidence metrics

## 🔮 Future Improvements

- Add support for image deepfake detection

- Implement multi-face detection and analysis

- Improve processing speed for longer videos

- Add detailed analysis of which parts of the video appear manipulated

- Develop a mobile application version

## 📄 License

This project is licensed under the [MIT](LICENSE) License - see the LICENSE file for details.

## 🔗 Links

[![linkedin](https://img.shields.io/badge/linkedin-0A66C2?style=for-the-badge&logo=linkedin&logoColor=white)](https://www.linkedin.com/in/balakrishna-kini?utm_source=share&utm_campaign=share_via&utm_content=profile&utm_medium=android_app)

⭐ If you find this project useful, please consider giving it a star on GitHub! ⭐