Adaptive three-dimensional histogram modification for JPEG reversible data hiding

Abstract

JPEG reversible data hiding (RDH) is a data hiding technique that requires both accurate data extraction and perfect recovery of the original JPEG image. Existing JPEG RDH schemes often rely on the distortion model of DCT coefficient frequency itself, failing to fully utilize the correlation between adjacent coefficients, resulting in inferior visual quality and significant file size expansion for JPEG image containing hidden data. To address the problem, we design a new JPEG RDH scheme by introducing three-dimensional (3D) histogram modification mechanism. We firstly evaluate the costs of each DCT block and frequency band to build coefficient triplet grouping mechanism. Furthermore, we construct a series of three-dimensional histogram mappings to perform data embedding according to the grouped DCT coefficient triplets, and then optimize the embedding efficiency by adaptively integrating multi-dimensional histogram mapping for the given embedding capacity. Extensive experiments demonstrate that our scheme significantly outperforms the state-of-the-art JPEG RDH schemes and can achieve efficient balance between higher visual quality and smaller file size changes while keeping JPEG file format unchanged.