Detection of Digital Tampering

Project Summary

The rapid increase in low-cost and sophisticated digital technology has made it remarkably easy to manipulate digital audio, image, and video. Coupled with vague legal definition as to the admissibility of digital media into evidence, it is essential that we develop techniques for authenticating digital material. We propose to: 1. Develop novel statistical models that embody the characteristics of "natural" signals/images. 2. Use these models in the development of techniques for determining the authenticity of digital signals/images.

Steganography: we have finished the porting of the steganography detection code from MatLab to C++, have fully tested this code and are working with United Devices to run this code on a massively distributed system. The goal is to find web pages that have suspicious activity.

Computer Graphics or Natural: we completed work on a new technique to distinguish between computer graphics and natural photographs. This work may be helpful in combating purportedly computer-generated child pornography.

Digital Tampering: we have completed development of three novel techniques for detecting various forms of digital tampering.

Voice Authentication: we have preliminary results that show that some of our image-based work extends nicely to audio. Specifically, we have developed techniques for distinguishing between natural and computer generated voices, and are working on extending this work to authenticate speakers from unscripted text.

• Project Lead: Hany Farid