EmbedR-Net: Using CNN to Embed Mark With Recovery Through Deep Convolutional GAN for Secure eHealth Systems

Abstract

The goal of security is to protect digital assets, devices and services from being disrupted, exploited or stolen by unauthorised users. It is also about having reliable media information available at the right time. However, the media distortion will pose many potential risks in eHealth systems. In this paper, a new data hiding method called EmbedR-Net based on Convolutional Neural Network (CNN) and Deep Convolutional Generative Adversarial Network (DCGAN) is proposed, which can prevent the copyright violation of the medical images. First, a CNN based embedder network is designed for imperceptibly hiding medical images as marks in the carrier image. Second, we compute the diagonal value of the marked image for hidden mark recovery. Last, the DCGAN network is designed to robustly recover the hidden mark using the diagonal value of the marked image. Compared to existing methods, experiments on five different datasets have shown that the proposed EmbedR-Net obtains superior performance.