Reversible watermarking by prediction-error expansion

Abstract

We propose a new reversible (lossless) watermarking algorithm for digital images. Being reversible, the algorithm enables the recovery of the original host information upon the extraction of the embedded information. The proposed technique exploits the inherent correlation among the adjacent pixels in an image region using a predictor. The information bits are embedded into the prediction errors, which enables us to embed a large payload while keeping the distortion low. A histogram shift at the encoder enables the decoder to identify the embedded location.