/real_vs_spoofed

A neural network that can discriminate between a live and a spoofed image, without needing explicit data of the spoofed images

Primary LanguageJupyter Notebook

Problem viewing files

If any of the ".ipynb" files don't load use nbvier to see them. Sadly github is known to have some problem with this file.

About this project

This is a program for my senior thesis. It consists of building a model that can discriminate between a real and a spoofed face. However without the necessity of the user giving any spoofed images to the training part of the model. For that, a GAN (Generative Adversary Network) that does style transfer will be used to the artificial creation of the spoofed images, using only real faces one, and just one for the creation of the desired style. The network used to create the poofed images is Fast Style Transfer.