Progressive Secret Sharing Through the Integration of Deep Learning and Reversible Data Hiding in Encrypted Images

Progressive secret image sharing (PSS) can progressively recover different resolution images or different parts of an image. To protect image privacy on the cloud and IoT application scenarios, this paper first presents an efficient PSS scheme through the integration of two deep learning networks and reversible data hiding in encrypted images. After sharing different parts of the original image, the receiver can progressively recover: a low-resolution image, a mask image where the most sensitive area is peeled off, and the original image according to the number of holding shares and image encryption keys. Some additional data such as signatures, fingerprints, and timestamps can be also embedded into each share on the cloud server, as the data extraction is separate from image recovery. Experimental results demonstrate that the proposed scheme can uniquely and efficiently obtain a recovered image progressively. Our scheme also has a higher security level, and the average embedding rate achieves almost a 1 bpp improvement compared to state-of-the-art schemes on a larger-scale image dataset.