Single-frame forensic watermarking for content authenticity

Robust technical solutions are crucial for reassuring the public about the authenticity of the information they consume and for providing organizations with the tools they need to actively combat the falsehoods that can harm their brands and reputations. The Coalition for Content Provenance and Authenticity (C2PA) is already making significant progress in this area by setting standards for certifying the authenticity and provenance of digital content through signed metadata. While this is a groundbreaking first step, we aim to enhance these efforts by integrating tamper-proof, robust watermarks that can withstand various transformations and ensure content provenance.

What is content provenance?

Provenance, from the French provenir (to come/stem from), refers to the historical origins of an object with the purpose of verifying its authenticity or ownership. It has traditionally been used in the art world to gauge the value of an object or entity, and in archeology and paleontology to distinguish the geographical origins of artifacts.

Similarly, digital content provenance or content provenance is used to mark the origin and evolution of digital materials, such as images, videos, audio files, and documents. Content provenance encompasses the intricate metadata that validates the authenticity and lineage of digital content. Within the digital landscape, digital provenance involves essential details, including the creator’s identity, the date of creation, its geographical origin, and any modifications made thereafter.

Recent calls from organizations such as the European Broadcasting Union (EBU) highlight the growing importance of industry-wide efforts like the C2PA to standarize content provenance across the media industry. These initiatives aim to establish a unified approach for ensuring content authenticity in an evolving digital landscape rampant with misinformation and deepfakes.

AI detection against deepfakes and misinformation

Deepfake technology has advanced rapidly, making it easier than ever to create convincing fake videos and images using AI. Initially a novelty, deepfakes have now become a powerful tool for misinformation, fraud, and manipulation.

Particularly, scams are becoming increasingly sophisticated with the aid of generative AI. The alarming incident involving an 86-year-old pensioner, who lost nearly $700,000 of his life savings due to an investment tip from a deepfake Elon Musk, vividly illustrates the severe repercussions that have emerged.

Facial features used to be the final obstacle in distinguishing between a genuine individual and a deep fake, but that barrier has been breached. The widespread utilization of StyleGAN2, an advanced neural network for producing faces, has facilitated the creation of counterfeit identities, enabling scams and the dissemination of misinformation. Likewise, various online platforms offer users the ability to craft personalized videos for fees as low as 10 US dollars.

But AI is a double-edged sword — both the threat and the solution. Several innovative Ai detection tools and platforms have emerged to tackle the challenges posed by deepfakes. Advances like MesoNet illustrate the potential of AI to make predictions on image data to identify real pictures from fake ones. Another notable example is Deeptrace, which employs AI algorithms to detect and trace manipulated media, offering comprehensive reports on the authenticity of content.

While AI detection is gaining popularity to combat scams and misinformation, there’s still a long way to go. Current deepfake detection relies on machine learning algorithms that analyze inconsistencies in pixels, shadows, and motion. These methods are constantly evolving, yet they struggle to keep pace with the rapid advancements in deepfake technology.

Single-frame forensic watermarking for content provenance and authentication

Castlabs has been at the forefront of innovating the watermarking space, developing our revolutionary single-frame forensic watermarking technology. This innovation, once deemed impossible by many in the entertainment industry, can detect an invisible watermark in a picture of a single frame of videos – where at least 15 seconds of leaked and unspoiled footage used to be required. And the same technology applies to images and photos as well.

Initially conceived to safeguard Hollywood’s content, our technology has evolved. We’ve created a proof of concept with the aim of not only leveraging the watermark to verify whether content originates from an authentic source but also tamper-proofing the content itself to enable determining if it has been altered.

The key to our technology is that the watermark can cover the entire real estate of the image, which enables detecting modified areas. With that information, we can create a heatmap highlighting areas that have been tampered with. Such tamper proofing, along with data that links to the original version, can be embedded in content published by news outlets, political entities, or corporations; serving as a stamp of authenticity directly from the source – see below for an example.

Benefits of single-frame watermarking

• Versatility: Our watermark technology is the only one capable of forensically watermarking both videos and images, surpassing competitors who specialize in one or the other.

• Single frame: The most advanced competitors in the market require at least 15 seconds of video to reliably identify a watermark, especially with degraded recordings taken with a smartphone camera – this is too long for short clips and does not work for a snapshot of a scene in a video.

• Invisibility and security: Unlike some authenticity marks that are visible and easily removable, ours is embedded imperceptibly into the content itself, making it significantly more resistant to tampering.

• Robustness: Unlike digital signatures, which can get lost during uploads to social media platforms, or simply removed by malicious actors, the Castlabs watermark is embedded within the content itself. This ensures it withstands recompression as well as a range of transformations offering a significantly more dependable solution for safeguarding the content’s proof of authenticity.

• Tamper-detection:This innovative feature detects any alterations, evidenced by disrupted watermarks, thereby providing forensic evidence of tampering.

How does it work?

As the industry standard, C2PA metadata serves as the first line of defense, offering users a quick and straightforward way to assess content authenticity where available. In most cases, this metadata remains intact, allowing users to easily verify that the content is genuine.

However, if the metadata is stripped away, the embedded watermark provides a second layer of protection. If users are unsure whether to trust content and want to verify its authenticity, they can use our cloud-based watermark extraction service, accessible via an app or website under the publisher’s branding. Users simply need to upload a screenshot or even a photo taken with their mobile device, and our service will quickly detect the watermark, enabling verification of the content’s authenticity and recovering of C2PA metadata.

Our service will confirm whether or not the content was published by the entity in question and whether it was originally C2PA certified but has since been altered – perhaps through voice cloning and face manipulation, or other AI-driven edits. With our tamper-detection technology, we can precisely identify which area of an image has been altered, whether it was a minor edit or a malicious alteration intended to spread disinformation.

With this forensic evidence, companies can flag suspicious content for further analysis and investigation. Additionally, they can offer users a side-by-side comparison of the modified content and the original, enhancing transparency and trust.

Single-frame watermarking against disinformation and deepfakes

This evolution of Castlabs’ single-frame forensic watermarking technology provides a tamper-proof, robust, and accessible tool to sit alongside C2PA metadata and equip the public with the means to confidently verify the authenticity of the information they encounter, ensuring what they consume is genuine and credible.

With growing concern over the potential impact of deepfakes and disinformation on the trust in institutions, public information, and the electoral process – manipulated content, mirroring the style and credibility of reputable sources, could go viral and significantly influence public opinion. In such critical times, the absence of robust and user-friendly technology capable of swiftly authenticating news content poses a real threat to the principles of a free and open public life.

Furthermore, we aim to provide companies with the reassurance that their brands and reputations are safeguarded against misuse by malicious entities. Our technology reduces the likelihood of the public being misled by disinformation presented under the guise of reputable branding, thus preserving the integrity of the company’s image.

The future of content authenticity

Ensuring content authenticity is an ongoing battle in a rapidly evolving digital landscape. Increasing demands from governmental bodies like the state of California reinforce the notion of forensic watermarking as a valuable tool to ensure provenance and content authenticity.

Single-frame forensic watermarking offers a pioneering solution for addressing the challenges posed by AI-generated disinformation and deepfakes. At Castlabs, we are committed to advancing the field of content authenticity and providing security professionals with the tools they need to safeguard their digital assets.