Reversible Data Hiding for Interpolated Images using CNN based Nearest Neighbour Interpolation

Interpolation based reversible data hiding (RDH) has gained plenty of research interest because to its ability to increase large amount of secret information in an image. However, one of the issues is that it involves approximating pixel values, which can result in a loss of image detail or in other words deteriorated image quality. To resolve this issue, this paper defines an innovative RDH method for interpolated images using convolution neural network based nearest neighbor interpolation (CNN-NNI). The CNN-NNI generates high-resolution image that closely matches the original image by processing through a series of convolutional layers. On the other hand, the strategy for embedding data involves using both interpolated pixels as well as reference pixels for data insertion in each block. Experimental findings indicate that the suggested approach beats recently developed and comparable techniques not only on final visual quality yet also at the maximum embedding ability i.e., higher than 1.75 bpp including an average PSNR value higher than 30 dB.