Gradient-based directional predictor for reversible data hiding

Abstract

Prediction-error expansion (PEE) is assumed to be the most efficient technique for reversible data hiding (RDH) recently. In PEE, generally, the data is embedded by modifying the prediction-error histogram (PEH). A sharper Laplacian distributed PEH can guarantee a larger embedding capacity (EC) and the accuracy of prediction on pixel is a key factor to form such PEH. In this paper, we propose a gradient-based directional predictor for PEE. It utilizes the gradient information of pixel's context and can generate more precise prediction result. Besides, in our method, the data is embedded in a pairwise manner for better marked image quality. The pixels are sorted into different collections according to their local complexities (LC) first and each group of pixels are optimally paired for data embedding. Experimental results show that our method outperforms some state-of-the-art techniques.