Blind Spread-Spectrum Steganalysis via Iterative Techniques

摘要

We developed theory and methods for optimal digital data hiding in arbitrary transform domains of digital hosts (images, video, audio). Our optimality criteria are: Host distortion, recovery error rate, and the Shannon capacity of the covert channel. Additionally, we introduced for the first time the concept of multiuser/multi-signature steganography. Then, we developed counter-measures to (optimal multiuser) steganography in the form of active (message extraction) and passive (stego/non-stego decision) steganalysis. We concluded that optimal data hiding, as described in the report, offers vast improvement in recovery error rate/Shannon capacity versus medium distortion and enables highly effective multi-signature embedding (different -potentially- hidden messages for different POCs along the chain of command etc.). In the context of active steganalysis, the developed M-IGLS hidden message extraction algorithm can destroy conventional SS steganography. However, our own optimal SS embedding is resistant to M -IGLS steganalysis attacks, especially for small hidden messages. Our new passive (binary hypothesis testing) steganalysis procedure offers close to 95% identification success rate at about 1 % false alarm rate when used on hosts with conventionally embedded messages.