Deep Learning-based Dual Watermarking for Image Copyright Protection and Authentication

TL;DR

The paper addresses how to simultaneously protect image copyrights and verify content authenticity in digital transmission by introducing a deep learning-based dual invisible watermarking framework. It embeds two distinct watermarks: a perceptual hash for copyright protection and a cryptographic hash for content and source authentication, with two independent extraction paths to avoid interference. The approach achieves high visual fidelity ($PSNR \approx 46.87$ dB, $SSIM \approx 0.94$) and strong verification performance (copyright ~$\approx 97\%$, authentication ~$\approx 95\%$), while exhibiting robustness to content-preserving manipulations and resilience against overwriting and surrogate-model attacks. This dual-watermark strategy enables automated, secure, and scalable protection of image copyrights and integrity in online ecosystems.

Abstract

Advancements in digital technologies make it easy to modify the content of digital images. Hence, ensuring digital images integrity and authenticity is necessary to protect them against various attacks that manipulate them. We present a Deep Learning (DL) based dual invisible watermarking technique for performing source authentication, content authentication, and protecting digital content copyright of images sent over the internet. Beyond securing images, the proposed technique demonstrates robustness to content-preserving image manipulations. It is also impossible to imitate or overwrite watermarks because the cryptographic hash of the image and the dominant features of the image in the form of perceptual hash are used as watermarks. We highlighted the need for source authentication to safeguard image integrity and authenticity, along with identifying similar content for copyright protection. After exhaustive testing, we obtained a high peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM), which implies there is a minute change in the original image after embedding our watermarks. Our trained model achieves high watermark extraction accuracy and to the best of our knowledge, this is the first deep learning-based dual watermarking technique proposed in the literature.

Figures (4)

• Figure 1: General overview of deep learning-based watermarking where an encoder network is used for embedding the watermark and the decoder network is used for extracting the watermark.
• Figure 2: Overview of the proposed Deep Learning-based Dual Watermarking. In the watermark embedding stage, two encoders, $E_1$ and $E_2$, are used to embed perceptual and cryptographic hash, respectively. There are two phases for the watermark extraction stage. In the first phase, decoder $D_1$ is used to verify the image ownership for copyright protection, while in the second phase, decoders $D_2$ and $D_3$ are used for content and source authentication. It is to be noted that both phases of watermark extraction are independent of each other.
• Figure 3: Visual quality of the image after embedding the watermarks using our technique.
• Figure 4: Perceptual hash of the image when different content-preserving image manipulations are performed.