This work optimizes an improved high-fidelity reversible data hiding scheme of Peng et al. which is based on improved pixel-value-ordering (IPVO) and prediction-error expansion. In Peng et al.'s method, the difference between the maximum and second largest value (or, the minimum and second smallest value) of a block is defined considering the pixel locations of maximum and second the largest value (or, the pixel locations of minimum and second the smallest value). When the difference between the maximum and second largest value (or, the minimum and second smallest value) of a block is equal to 0 or 1, the block can be exploited to embed data. Otherwise, the block should be shifted or remain unchanged. However, different prediction-error used to embed information can lead to different histogram modification and different pixel shift rate, to further reduces the change in the carrier image. In this work, we list all the different prediction-error, which are used as the selection object for the embedded error when hiding information. As a prerequisite of meeting the demand of the embedding capacity, some appropriate prediction-errors are selected for embedding to reduce the number of the pixel shifted in the marked image as small as possible. An IPVO-based reversible data hiding scheme with floating predictor is also extended. Experimental results show that the proposed scheme yields a superior performance than the state-of-the-art works, under the condition of same embedding capacity, especially for relatively rough images.
Keywords: PSNR; Reversible data hiding; floating prediction-error; improved pixel-value-ordering; optimization.