A novel two-level embedding pattern for grayscale-invariant reversible data hiding

Although traditional reversible data hiding (RDH) methods work well at protecting the secret data and guaranteeing the high visual quality of the cover image as well, the performance of down steaming visual tasks, such as feature extraction and point alignment, is degraded. To address this issue, Hou et al. [13] proposed to embed secret data into color images while keeping the corresponding grayscale invariant. However, we find that the embedding efficiency of this method is still unsatisfactory. And the embedding distortion should be further reduced. In this paper, unit embedding distortion is designed to help evaluate the embedding efficiency of per pixel unit. Then, an adaptive embedding pattern is introduced by adaptively embedding one or two bits of secret data into different positions according to the context information. While our approach provides a novel two-level predictor, benefiting from two normal predictors to reduce the embedding distortion. Experimental results demonstrate that, compared to the previous method, our scheme could significantly enhance the image fidelity while keeping the grayscale invariant.