Reversible Data Hiding Based on Elastic Net Predictor

Recently, researchers in the information security field have shown increasing interests in reversible data hiding (RDH). Two key aspects of RDH are accurate image prediction and minimizing distortion during embedding. This paper proposes a novel RDH method based on the elastic net predictor. The elastic net is a penalized least squares algorithm that addresses the overfitting problem by minimizing the sum of residual squares (RSS) with a joint constraint on the L1 and L2 norms of the coefficients. The proposed method divides the cover image into two subimages using a rhombus pattern. The elastic net predictor was employed to accurately predict the pixel values of each subimage. Secret message was embedded into the subimage by using the prediction error expansion-histogram shifting (PEE-HS) scheme, leading to a further distortion reduction. Experimental results demonstrate the superiority of the proposed method over existing RDH schemes in terms of prediction accuracy.