Reversible Data Hiding in Encrypted Images Based on Quantization Prediction Error

Abstract

Reversible data hiding in encrypted images (RDH-EI) can simultaneously protect secret data and the content of transmission carriers, so it has important applications in cloud computing, medicine, and other fields involving data privacy. Aiming at the problem of low embedding capacity in current RDH-EI algorithms, a separate algorithm based on interpolation prediction error quantization is proposed. First, we down-sample the cover image to obtain the sampled pixels. Then interpolate and predict non-sampled pixels to obtain auxiliary data such as prediction errors and classify the auxiliary data. By introducing a quantitative prediction error loss factor, auxiliary data can be compressed to various degrees, reducing the amount of auxiliary data. Next, auxiliary data is embedded into sampled pixels through reversible data hiding (RDH) technology, and the partial of non-sampled pixels. Finally, according to the hiding key, the mark data and secret data are embedded in the encrypted image. At the receiver, a legitimate user extracts secret data and recovers the cover image with owned keys. Experimental results show that the proposed algorithm can ensure the error-free extraction of secret data and provide lossy and lossless versions of the cover image. In lossy versions, the maximum embedding rate can be around 4bpp. Compared to other advanced algorithms, the proposed algorithm has the advantages of more flexibility, a high embedding rate, and recovery quality.