Reversible Data Hiding Using Multi-Layer Perceptron Based Pixel Prediction

Abstract

Reversible data hiding is a branch of data hiding in which cover image can additionally be restored along with the extraction of hidden data. In prediction error expansion based reversible data hiding techniques, pixel values are predicted using the neighborhood pixels. Data bits are hidden in the expanded prediction error. Prediction error expansion based techniques have become popular due to their superior performances over other categories of techniques. In this paper, a novel reversible data hiding technique is proposed based on a multi-layer perceptron based pixel prediction. The multi-layer perceptron is trained to predict a pixel value based on its eight neighbors. Experiments have been conducted by varying the number of hidden layers and the number of neurons in each hidden layer in the multi-layer perceptron. The best architecture for the multi-layer perceptron is chosen for the proposed reversible data hiding technique. Finally, an adaptive embedding strategy is used to insert either one bit or two bits of data in the expanded prediction error of the pixel based on the standard deviation of the pixel values in the local neighborhood.