Filename Ballistics | Given a filename, identify the type of camera or online service that generated the picture. The current solution is applicable over 60% of the time, and averages over 95% accuracy. |
---|---|
Prohibited Content Detection (aka "anti-porn filters") | The public FotoForensics site prohibits pornography, nudity, and sexually-explicit content. This research project is designed to rapidly identify prohibited content. The current solution can be generalized into any kind of rule-based automated content detection. |
Trend Detection | Identify news before it is reported as news. |
Fraud Detection | Fraud using altered photos for banking, insurance, and other services is a multi-billion dollar industry. Can we detect the tools and techniques that the professional fraud groups use?
|
Precog | Most users just want to use the site, but some want to attack. Can we reduce our attack surface or identify attackers before they attack?
|
Similar Picture Search | Given a picture, quickly find other visually-similar images. We are developing a variety of picture search engines. The fastest so far is also one of the best: it can compare nearly 3 billion signatures per second (single-threaded and without special hardware) -- nearly 3000x faster than PhotoDNA (on the same hardware) and just as accurate.
|
Forensic Tool Usability and Simplification | Digital photo forensics is a complicated topic. What can be done to lower the learning curve and increase accuracy for the analyst?
|
Data Visualization | How do you visualize results and identify trends? |
Network Distribution | People have been talking about IPv4 address exhaustion and the migration to IPv6 for decades. But how many systems actually use IPv6? Are there any common issues related to IPv6? |
Browser Profiling | Different browsers support different features. If we want to support the most common browsers and service providers, then what functionality should we support? |