Bayesian structural content abstraction for image authentication

An ideal authentication approach should be able to tolerate "content preserving" operations (e.g. compression and rotation) robustly, while detecting "content altering" (e.g. object removing and replacement) modifications sensitively. Although the numerous existing algorithms can detect and locate malicious alterations to the protected image, unfortunately, most are still too fragile to resist various content-preserving manipulations for enhancing the sensitivity to content tampers. In this paper, we propose a structural content abstraction scheme within the Bayesian framework, based on which an implicit and reliable image authentication scheme can be established. Because BSCA provides a trustworthy reference which correlates highly with the content, the proposed authentication method possesses both robustness to non-content operations (NCO) and sensitivity to content operations (CO). Experimental results also