Although steganography - the practice of hiding a secret message within an innocuous cover medium - is hardly new, detecting the presence of hidden messages has become considerably more difficult with the advent of high-resolution digital images as carriers.

We have been developing statistical techniques to detect (but not necessarily decipher) the presence of hidden messages within a digital image. The basic approach works by finding predictable higher-order statistics of "natural" images within a multi-scale decomposition, and then showing that embedded messages alter these statistics.

In related work, we have developed techniques to differentiate between photographic and photorealistic images, and to detect traces of digital tampering.

Matlab routines

(Collaborative work with Kimo Johnson and Siwei Lyu)