Fake image detection based on fact checking
Manipulated videos, audio and images have been around for years, but the increase in access to photo apps in recent years have made it easier for anyone to create fake images/videos. Meanwhile, the rise of artificial intelligence and advanced editing software have made them much harder to spot. O...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2024
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/176725 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Manipulated videos, audio and images have been around for years, but the increase in access to photo apps in recent years have made it easier for anyone to create fake images/videos. Meanwhile, the rise of artificial intelligence and advanced editing software have made them much harder to spot.
One form of fake image and videos is to use old and unrelated images as current photos. This is identified as the main form of fake content in the recent Russia-Ukraine information war. For example, a gas explosion in the Russian city of Magnitogorsk in 2018 has been claimed as images and clips from the destruction of a residential building in Chuhuiv in eastern Ukraine.
In this project, the author aims to leverage on machine learning, specifically Convolutional Neural Networks (CNNs) to analyse diverse datasets and approaches to address out-of-context images. By investigating various pre-trained models, finetuning their performances and conduct training using supervised learning techniques, as well as evaluating their performance in identifying out-of-context images. Lastly, the author also attempts to propose software architecture and implement using these models to provide a proof-of-concept in real-life application for combating out-of-context fake news.
The project emphasises the importance of detecting out-of-context images using multiple approaches, with a focus on exploring neural network architecture and training techniques that may facilitate fake news detection. It facilitates in laying the groundwork for both future research and application for development and deployment of more robust and accurate image forensic machine learning models. |
---|