Multiple predictors based RW scheme with adaptive image partitioning

Abstract

Reversible watermarking (RW) is one of the best possible solutions for content authentication of a digital data. In RW the decoder may recover the hidden and original information losslessly. Existing works suggest that prediction error expansion (PEE) based RW scheme ensures higher embedding capacity with low imperceptibility. In general, PEE based RW schemes use a single predictor. But this is seen that the different types as well as different regions of an image behave in different way during embedding. So, this work presents a RW scheme based on local characteristics of an image, where multiple predictors are used to enhance the embedding bit rate. To this aim, an image is partitioned into smooth, texture and edge regions using adaptive threshold values. The threshold values are calculated by maximizing the fuzzy conditional entropy of the gray values; where the optimal set of parameters for the fuzzy membership functions is specified by differential evolution method. A large set of simulation results are shown to highlight its improved rate-distortion performance over the existing works.