DeceptiGuard

Decptiguard
Short Description

Inspiration

With AI on the rise, there seems to be no end to misinformation. The internet can be a scary place, especially with Deepfakes and AI making it even scarier. So, we put on our thinking caps and found a way to tackle a part of this problem in realtime.

What it does

DeceptiGuard is a deepfake detection tool designed to identify Deepfake media content in realtime with relatively good accuracy. It utilizes Convolutional Neural Network (Mesonet) and Machine Learning algorithms to analyze images and videos, flagging suspicious content that may be a result of deepfake manipulation. By providing real-time detection capabilities, DeceptiGuard helps users discern between genuine and DeepFake media.

TLDR; If it detects you watching a deepfake news or media it alerts you!.

How we built it

Here are the following tools we used:

Tensorflow for Machine Learning
Database used to train: https://e.pcloud.link/publink/show?code=XZnsxkZkEAgI1OgQIJHLnNl9ErhV4vpHuV0
OpenCV library to monitor the computer screen for faces and process the faces
Mesonet for Prediction
Win11toast library for Windows notification

Challenges we ran into

While training the our model we could not use high quality deepfake images to train our model because analyzing patterns in high res images would require immense computing power and time. Also, Based on a recent study, "The State of Deepfakes" conducted by Deeptrace Labs, it was found that 90% of deepfake content is of pornographic nature. So, sourcing a substantial number of non-pornographic deepfake images was challenging. Anyways, we were able to find some databases from kaggle and other sources.

Accomplishments that we're proud of

Initially, our idea was to create a website where users could upload an image, and the website would determine whether it was a deepfake or not. However, as we progressed with the development, we are pleased to have built a system capable of real-time detection for various types of images and videos. We are also proud that our system shows relatively high accuracy in distinguishing between real and deepfake media.

What we learned

OpenCV Tensorflow Dangers Of DeepFake Teamwork Planning

What's next for DeceptiGuard

We're thinking of expanding our scope of detection to encompass audio deepfakes and text-based manipulation, integrating real-time monitoring capabilities for social media platforms and news outlets, and exploring opportunities for collaboration with industry partners and cybersecurity agencies. Ultimately, our goal is to empower users with the tools and knowledge needed to combat the evolving threat of deepfake technology effectively.

(Maybe pitch this idea to Microsoft to add it in their next windows update 🤣🤣🤣)