DEEPFAKE
DETECTOR
PYTHON
HYBRID AI
FORENSICS
THE_OBJECTIVE
With the rapid advancement of GANs, hyper-realistic fake videos have become a severe threat. Traditional forensic methods fail to detect high-quality face-swaps. The goal was to build a system capable of catching these forgeries by analyzing temporal inconsistencies.
CORE_SYSTEM_LOGIC
Instead of relying purely on spatial frame-by-frame analysis, I engineered a Hybrid Neural Network. The system uses CNNs (like ResNet) to extract facial features, and feeds those into an RNN (LSTM) layer to analyze unnatural flickers and movements across a sequence of frames.
UX_EXECUTION
The backend pipeline processes video inputs using OpenCV to isolate the face, runs the sequential prediction via Keras/TensorFlow, and outputs a confidence score indicating the probability of manipulation.
TECH_ARSENAL
PYTHON
TENSORFLOW / KERAS
OPENCV
CNN + RNN (LSTM)
MTCNN