Image-Caption Geolocation for Privacy

This project hopes to analyze the text & image content in a social media type of post to find locality-based information in order to identify sensitive information about an individual. It warns an individual about the potential information leak and indicates ways to block it. This project resulted from an paper from coursework.

The concept is to help prevent information leaks that can happen without people even knowing. With the amount of public data available about people, it can be simple to identify anonymous user accounts to their more personal ones. This project is an initial attempt to address those issues with a growing social media sphere and a rising number of users attempting to remain anonymous. Especially on sites like Reddit, where users may think they are completely anonymous, their identity and/or personal information can easily be discovered.

This tool was developed for a Computer Vision and Language class with the help of fellow students: Nick Mohammad and Ramya Bhaskara.

Here is a demo video illustrating what our tool can do. Unfortunately, we cannot post this tool as a website for everyone to use due to computing constraints (and none of us own a server).

The source code for the demo is available on GitHub.