Prediction-based reversible image watermarking using artificial neural networks

Abstract

In prediction-based reversible watermarking schemes, watermark bits are embedded in the prediction errors. An accurate prediction results in smaller prediction errors, more efficient embedding, and less distortion for the water- marked image. In this paper, an accurate prediction is made using artificial neural networks. Before the embedding operation, 2 neural networks are trained by the pixel values of the image. Then the trained neural networks predict the pixel values that are used in the embedding operation. Due to the training ability of the neural networks, the prediction will be more accurate than the averaging technique. Experimental results show that the proposed scheme yields superior results compared to several related schemes.