Reversible image authentication scheme with blind content reconstruction based on compressed sensing

Abstract

Reversible image authentication (RIA) schemes embed fragile watermarks to protect images against tampering attacks. If a marked image is not attacked, watermarks can be completely extracted and eliminated to obtain the original cover image. Tampered image regions can be reconstructed by including recovery information as part of the watermarks. In this paper, we introduce a novel blind RIA scheme for tampering localization and reconstruction. The proposed scheme is based on compressed sensing technique to obtain reference values, which are embedded into the cover image to perform sparse signal reconstruction using only the embedded information. Signal sparsity is enhanced in the discrete cosine transform (DCT) domain by omitting some high-frequency coefficients. Experimental results indicate that better sparse signal approximations can be obtained when some DCT coefficients are omitted, improving image reconstructions. The proposed scheme outperforms similar state-of-the-art works when reconstructing tampered image regions, especially when the damaged area size is increased.