Minimizing additive distortion functions with non-binary embedding operation in steganography

Most practical steganographic algorithms for empirical covers embed messages by minimizing a sum of per-pixel distortions. Current near-optimal codes for this minimization problem [7] are limited to a binary embedding operation. In this paper, we extend this work to embedding operations of larger cardinality. The need for embedding changes of larger amplitude and the merit of this construction are confirmed experimentally by implementing an adaptive embedding algorithm for digital images and comparing its security to other schemes.