Reversible data hiding based on prediction-error value ordering and multiple-embedding

Abstract

Prediction-error expansion (PEE) is the most widely investigated reversible data hiding (RDH) framework. However, the performance of PEE can be further improved since the image redundancy has not fully exploited yet. To enlarge the embedding capacity and optimize the embedding performance of RDH, in this paper, a novel high-fidelity reversible embedding scheme using prediction-error value ordering and multiple embedding is proposed. Incorporating with PEE, the inter-correlation of prediction-errors is exploited by ordering the prediction-errors values, and multiple to-be-modified prediction-errors within each image block are adaptive determined. Specifically, firstly, a newly defined prediction-error is computed by considering the difference between the top two maximum (minimum) prediction-errors within each image block. Then, these prediction-errors are modified for expansion embedding. Moreover, a multiple-embedding mechanism is introduced to adaptive determine the number of modified prediction-errors. Finally, extensive experiments are conducted, and it is experimentally verified that the proposed method is superior to than some state-of-the-art RDH works. A benchmark result is that, for the standard Lena image, a PSNR as high as 62.01 dB can be derived for an embedding capacity of 10,000 bits.