Reversible data hiding based on novel embedding structure PVO and adaptive block-merging strategy

Abstract

In data hiding field, pixel-value-ordering (PVO)-based methods yield excellent performances by taking full advantage of correlation in a fixed block. In PVO-based methods, the maximum and minimum of each block are utilized to embed data. The maximal embedding capacity (EC) is achieved by setting embedding block size smaller to generate more basic embedding blocks. In this light, a new three-pixel embedding structure is proposed to improve the maximal EC by reducing the number of pixels in each basic block. Meanwhile, unlike conventional algorithms which take the noise level of current block as complexity, we utilize the correlation information among neighboring blocks to get a better classification on embedding blocks. Moreover, dynamic block partition method is further investigated and an adaptive block-merging strategy is proposed in this paper. In this strategy, a more rational way is provided to exploit the embedding potential than the top-down partition method and the performance is significantly enhanced. With these three new improvements, our scheme exhibits better performance than the state-of-the-art PVO-based methods.