Dynamic content selection-and-prediction framework applied to reversible data hiding

The existing reversible data hiding (RDH) methods often use a fixed pixel preselection pattern and predictor to generate prediction errors that are then utilized for embedding secret data. According to Kerckhoffs's principle, this deterministic operation may allow an illegal decoder to successfully reconstruct the marked prediction-error histogram from a marked image, which is not desirable in application scenarios. This has prompted us to propose a dynamic content selection-and-prediction framework for the RDH in this paper. The proposed framework aims to auto-preselect the complex pixels out from a given image to predict the rest pixels that are thereafter exploited to carry the secret data. Comparing with some state-of-the-art algorithms, the proposed technique guarantees that, the illegal decoder will hardly locate the whole marked pixels and determine the marked prediction errors, which can ensure the security level. In our designed framework, after pixel selection and prediction, there exists a lot of freedom to design the data embedding procedure, meaning that, the proposed framework can be applied to the design of an RDH scheme. In the experiments, we simply employ an optimized histogram shifting operation for data embedding after applying the proposed framework. Our experimental results have shown that, the data embedding process can benefit from the proposed pixel selection and prediction procedure with relatively low embedding rates, and therefore significantly outperform some related works in terms of the payload-distortion performance, especially for images with more smooth regions.