Prototype Description

Veracity AI at the Florida State University College of Communication and Information, founded by Dr. Shuyuan Metcalfe, is committed to automating the detection of manipulation in images using artificial intelligence. Our research assists in the fight for truth by efficiently identifying fabrication in visual media with accuracy. 

The prediction architecture uses the deep consensus algorithm running on an on-site set of high-performance computing nodes. The deep consensus model generates a prediction mask, indicating where manipulations occur in the image on a pixel level. The model is trained in three manipulation categories: splicing, inpainting, and copy-move. 

Splicing is the process of inserting a region of an image into another image. In the example below, the red and white plane is spliced into the image.

Inpainting is the process of removing a region of an image. In the example below, a bus on the left is removed using inpainting on the image. 

Copy-move is the process of copying a region of an image and pasting it into another region of the image. In the image below, the hotdog with mustard is copied and moved to the top of the image.

When a user logs in on our website, they will be able to submit images to the prediction architecture as well as access their submission history. Once submitted, images enter a processing queue for the model. Users can expect their submission to receive predictions in around 2-5 minutes from the time they were successfully uploaded. The prediction results can be accessed in a user’s submission history, where the user’s image, prediction, and confidence score for each submission can be found.