PEE based RW using fuzzy conditional entropy for image partitioning

Abstract

Prediction error expansion (PEE) based reversible watermarking (RW) has found to be efficient for meeting the high embedding rate at low visual distortion. However, the existing works mostly use single predictor over the entire host image. Further performance improvement is possible using predictors based on local characteristics of the image. To this aim, this work first proposes a method to partition the image into different regions, namely the smooth, the texture and the edge regions using multiple thresholds on pixel gradients. The threshold values are calculated by maximizing the fuzzy conditional entropy of the gradient values. The optimal set of parameters for the fuzzy membership functions are specified by differential evolution method. Two predictors are then proposed, one for prediction of gray values in the edge region and the other one for the texture and the smooth region. RW is then done using region specific PEE. A large set of simulation results are shown to highlight its improved rate-distortion performance over the existing works followed by semi-fragile nature of watermark decoding against common operations like smoothing filtering, noise addition, cropping, random bending attack, etc.