General distortion based reversible data hiding for JPEG images

Reversible data hiding (RDH) for JPEG images plays a vital role in image authentication and secure visual communication. However, most existing methods assign uniform distortion costs to all embedding positions, overlooking the fact that modifications in smooth regions are more perceptible to the human eye. This leads to noticeable visual artifacts. To address this, this paper proposes a novel general distortion–based RDH framework that introduces an adaptive distortion model tailored to the human visual system, enabling more perceptually consistent embedding. Specifically, a position-aware distortion model that adaptively weights each DCT coefficient is developed to prioritize modifications in visually complex regions. Then, matrix embedding that minimizes embedding distortion is employed to transform secret data into low-impact modification sequences. By prioritizing complex regions and penalizing smooth areas, the proposed method significantly reduces visible artifacts in marked images. Experimental results show that the proposed method effectively reduces the modifications on smooth areas, and achieves better performance in terms of PSNR and SSIM in comparison with several classical and state-of-the-art RDH methods. These results demonstrate the potential of perceptual modeling in advancing RDH performance and encourage further research in this direction.