Digital image watermarking using deep learning

Abstract

At present, watermarking techniques play an important role in protecting digital images. To date, many classical watermarking schemes have been developed to protect images based on spatial and transform domains. However, classical watermarking schemes are less resilient to many attacks. Recently, deep learning-based watermarking made a significant contribution to image content security and received attention for various popular applications. In this paper, we use convolutional neural networks (CNNs) to propose an interesting watermarking technique for digital images. Initially, latent features of cover and secret images are extracted using an encoder network and later concatenated to generate a marked image. On the receiver side, a denoising autoencoder network is used to remove noise variations from the received image and later to extract the secret mark image using a CNN. Our technique not only imperceptibly hides an image inside a cover image but also outperforms other state-of-the-art schemes in terms of visual quality and robustness according to simulation results and performance comparisons.