摘要

Constructing the auxiliary data is important in some reversible data hiding (RDH) techniques, where both the auxiliary data and the secret data are embedded into the cover image. The auxiliary data can be extracted and served as side information for the image recovery. In this paper, we propose a novel auxiliary data construction scheme for RDH in JPEG images. In order to preserve the file size, we adopt the RDH framework proposed by Fridrich et al., where the least significant bits (LSBs) of the original AC coefficients (termed as the original LSBs) are compressed directly as the auxiliary data. In order to achieve a higher compression rate, we propose to reconstruct the original LSBs from the corresponding AC coefficients with all the LSBs set to zero, which is performed by analyzing the block artifacts in the spatial domain. The auxiliary data is then constructed by compressing the original LSBs conditioned on the reconstructed LSBs. In conjunction with the auxiliary data construction, we also propose a coefficient selection strategy to further improve the compression rate. The experiments show that the size of our auxiliary data is smaller than that compressed from the original LSBs directly. In addition, the RDH framework with the proposed scheme outperforms the state-of-the-art JPEG RDH techniques with file size preservation.