Data scientists create tool to spot fake images

Pixelator v2 is a tool to spot fake images. It uses a new combination of image veracity techniques with capability beyond what can be seen by the human eye. It can identify subtle differences in images with greater accuracy than traditional methods and has been shown to detect alternations as small as 1 pixel.

Highlighting differences between distorted Lenna and reference Lenna images using SSIM and Pixelator v2

The tool is developed by York St. John University’s Data Science academics, supported by colleagues from the University of Essex and software developers at Nosh Technologies.

Spot fake images with accuracy

Pixelator v2 is designed to support those with the greatest need for accuracy, and the team says the software will be of particular use to cybersecurity professionals, analysts, and researchers. It employs an advanced method that combines:

LAB Colour Space Analysis: A perceptual color model mimics human vision, allowing the tool to detect differences in images that may not be immediately visible to the naked eye.

Sobel Edge Detection: A technique that highlights structural variations in images, such as changes in edges and boundaries.

Download

The research team is actively working on the next phase of this project, extending the tool to detect and predict generative AI-created images directly.

Pixelator v2 is available for free on GitHub.