Reversible demosaiced image authentication scheme with recoverability using clustering and matching techniques.

In this paper, we propose a reversible and recoverable authentication technique for demosaiced images using the k-means clustering algorithm and side match technique. Prior methods generate the recovery codes directly using the most significant bits of sampled components or their compressed codes, which requires excessive bits to record them. This not only damages the image to a large extent, but also limits the quality of the recovered image. In this paper, the sampled components are clustered and their cluster indices are used as the recovery codes. A tampered block can mostly be recovered by other untampered ones with the same cluster index. For an unrecoverable block, the cluster indices of its neighboring untampered blocks are employed to predict the block, and the recovery is performed by using the side match technique. Since unrecoverable blocks are sparsely scattered in the tampered areas, the recovered blocks around them are mostly available, and therefore an accurate prediction can be made. The experimental results show that the average quality of marked and recovered images of our method increases by around 3.36 dB and 2.96 dB, respectively, compared to previous methods. [ABSTRACT FROM AUTHOR]