SINR, bit error rate, and Shannon capacity optimized spread-spectrum steganography

For any given host image and (block) transform domain of interest, we derive the signature vector that when used for spread-spectrum (SS) message embedding that maximizes the signal-to-interference-plus-noise ratio (SINR) at the output of the maximum SINR linear filter receiver. Under a (colored) Gaussian assumption on the transform domain host data, we see that the same signature offers minimum probability of error message recovery at any host distortion level or -conversely- minimizes the host distortion for any probability of error target level. In addition, we show that the same signature maximizes the Shannon capacity of the covert link. All developments are then generalized to cover SS embedding in linearly processed block transform domain host data with orders of magnitude demonstrated improvement over current SS steganographic practices.