Secret and public key authentication watermarking schemes that resist vector quantization attack

In previous work, Wong had proposed both secret key and public key watermarking schemes for image authentication that can detect and localized any change made to a watermarked image. The techniques proposed were block-based, that is, they partitioned the image into non-overlapping blocks and separately authenticated each block. Subsequently, Holliman and Memon observed that many block based watermarking schemes are vulnerable to substitution attacks. They specifically showed that the Wong schemes can be attacked using a `vector quantization' (VQ) approach. This attack exploits that fact that if a sufficient number of images containing the same watermark bitmap is available, then one can use a VQ-like technique to forge a watermark into a new image. About the same time and independently, Coppersmith et al. proposed to use overlapping blocks to resist this attack. Although this method can make the attack inefficient, it does so with a significant loss of the localization property of the watermark. We extend in this paper the Wong schemes so that the resulting algorithms can resist the VQ attack and at the same time provide the same localization property in the watermark as the original schemes. The key idea is to insert a unique image-dependent block ID into the watermarking process so that the VQ attack will not have a rich enough `codebook' to forge the watermark.