Extraction of Spread-Spectrum Hidden Data in Digital Media.

摘要

This paper considers the problem of blindly extracting data embedded over a wide band in a spectrum (transform) domain of a digital medium (image, audio, video). We first develop a multi-signature iterative generalized least- squares (MIGLS) core procedure to seek unknown data hidden in hosts via multi- signature direct-sequence spread-spectrum embedding. Neither the original host nor the embedding signatures are assumed available. Then, cross-correlation enhanced M-IGLS (CCM- IGLS), a procedure described herein in detail that is based on statistical analysis of repeated independentM-IGLS processing of the host, is seen to offer most effective hidden message recovery. Experimental studies on images show that the proposed CC-MIGLS algorithm can achieve recovery probability of error close to what may be attained with known embedding signatures and host autocorrelation matrix.