Reversible image watermarking on prediction errors by efficient histogram modification

A reversible data hiding algorithm is proposed, in which the efficiency of modifying a pair of histogram bins is considered. Multiple pairs of histogram bins can be further selected for data embedding in sequence, while pre-process of pixel values is performed to prevent the possible overflow and underflow. Embedding with the prediction errors is investigated with a new prediction scheme. In each of the four prediction modes, a large amount of prediction errors can be produced from the host image. Moreover, all combinations of the four modes to generate a number of histogram pairs are enumerated to obtain the best performance. Blind extraction and recovery are enabled by embedding a pre-computed location map and other overhead information into the watermarked image. Promising experimental results are obtained on a variety of test images. Compared with the existing algorithms, the image content is better preserved in high payload data hiding.

► Efficiency of data embedding is defined for the histogram modification algorithms.
► We consider the defined efficiency in selecting the histogram bins to be modified.
► A large number of prediction errors are produced with a new prediction scheme.
► Performance of high payload embedding is improved by multi-pair embedding.
► The optimal prediction mode is chosen by enumeration and performance comparison.