Pixel Complexity Sorting Embedding for Reversible Data Hiding Based on Elastic net Predictor

Reversible data hiding techniques have increasingly garnered attention from researchers in the field of information security due to their capacity to recover the original image non-destructively and their substantial embedding capacity. Over recent years, researchers have made strides in achieving commendable prediction accuracy through the use of linear regression models. However, it is important to note that linear regression models such as ridge and lasso regression have limitations and their performance is not always optimal. In this paper, we propose a novel reversible data hiding model based on an elastic prediction network and pixel complexity order. The elastic net predictor combines the advantages of ridge and lasso regression, incorporating L1 and L2 paradigms as penalty terms. This paper divides the image into a dot set and a cross set, and the elastic network trains the rhombus predictor. Subsequently, it is segmented into a two-stage prediction process, with embedding occurring in the order of pixel complexity from low to high. The proposed method outperforms other existing linear regression predictor methods at low embedding loads.