Universal lossless data embedding without explicit compression

In digital watermarking, the original data, called host or cover data, is inevitably distorted by the insertion of a watermark or embedded data. However, there are a number of applications that require complete lossless recovery of original data. This paper introduces an information theoretical model to the lossless data embedding problem, and gives the maximum embedding rate of the model. The main contribution of the paper is to propose an embedding scheme, which can asymptotically attain the maximum rate without knowing the source of the host data. The proposed scheme divides the host data into fixed-length blocks and embeds a watermark by replacing a less significant component of each block with a new component of the same length. The new component represents both the replaced original component and a fragment of the watermark. This paper shows that as the block length tends to infinity, the embedding rate converges to the maximum rate that could be attained when we ideally compress the less significant components using complete knowledge on the source of the host data.