Trusted media capture
There is more to our images (and videos) than meets the eye.
Up to 500 anonymised data points per photo or video
Trusted Media Capture™ is the only solution on the market which does not upload and store photos and videos to the verification provider’s servers. Instead we poll 300-500 data points from a photo/video at the point of capture. These data points are linked only to the media item, and never to a named user.
Under 15kb per capture
Unlike competing solutions which typically transfer anywhere between a few to dozens of megabytes off the device to document a capture event. Trusted Media Capture™ transfers under 15kb – resulting in a nimble solution requiring minimal mobile bandwidth, one that can work even with adverse web connectivity and with minimal drain on the mobile device’s battery.
Thousands of checks to date
We’ve run thousands of verification checks relating to content and metadata authenticity to date.
Serelay photos and videos have been captured by users in over 50 countries, across 6 continents.
Fundamentally, we facilitate the truth
We’ve founded Serelay based on two observations – the first is what we call ‘retroactive’ verification – just getting a photo or video off the web and trying to verify it is often not viable. Digital forensics were built for one analyst examining one photo over an extended period and were never intended for the scale of modern photo and video usage. Moreover, synthetic media and Deepfakes (AI-generated media) make this process continuously more challenging.
Our second observation was more optimistic – looking at the capabilities of a modern mobile device (computing capacity, sensor array, security infrastructure) we can, and should, enable mobile device users to capture photos and videos in a way which is inherently verifiable: point-of-capture verification.
A different approach
Serelay is not the only point-of-capture verification solution on the market, but it is decidedly unique.
Our Trusted Media Capture™ technology is the first commercial solution eschewing the chain-of-custody approach currently prevalent in the market.
While chain-of-custody solutions rely on transferring photos and videos in their entirety to a verification-provider’s servers as soon as they are capture, we do not. Instead, Serelay has developed an Integrity Vector approach – the computation of over 100 mathematical attributes relating to the media file at the point of capture, and it is only these values, alongside rich sensor metadata which can be securely transferred off-device.
Using only this integrity vector, Trusted Media Capture™ can provide an assessment of the veracity of the corresponding photo/video to detect a single pixel/frame change in content, change-level-analysis (to what extent has the photo been altered? Which are the affected regions?), photo-of-photo detection (also referred to as ‘re-broadcasting’) and physical spoofing of sensory input (such as GPS-signal spoofing and mocking).
Privacy by design
As opposed to the direct-chain-of-custody approach which entails storing any photo or video immediately after capture (regardless of whether or not the user chose to upload them), no photos or videos are stored on Serelay’s servers – only a list of mathematical values.
This privacy by design approach means no media can be viewed by Serelay staff, leaked from Serelay’s servers, or hacked. The mathematical values themselves are never attributed to a named user, nor can they be used to re-create the media item they support, and are hence of no material value on their own.
Best in class mobile bandwidth utilisation, battery utilisation and user experience
Transferring a data load of less than 15kb per captured media item – regardless of that media item’s size– means that even with a weak/unreliable Wi-Fi or mobile data connection, Trusted Media Capture™ can be provided, and a user’s mobile data allowance goes virtually unaffected, and battery drain is kept minimal.
Stopping users from getting around Trusted Media Capture
Photo of Photo detection: An “analog hole” or “rebroadcast attack,” are common terms for subverting provenance systems by capturing an image of a photograph or computer screen. By detecting the presence of flat surfaces in the image or video, we are able to classify whether the picture is of a real-life event or an image of a photograph or computer screen.
Change level analysis: A malicious user can inject, remove or modify objects from the image either at the point of capture or after the capture. Technologies based on cryptography and computer vision help detect to what extent a photo is altered and what are the affected regions.
Image injection detection for detecting if the desired image is injected/overlayed into the mobile device’s visual input stream.
Physical spoofing of sensory input: While mobile operating systems provide application access to a device’s location (pending a user’s permission to do so), the location itself can be altered by either physically spoofing the location signal received by the device or, a far more prevalent approach, by circumventing the operating system’s software. We have modelled an algorithm to detect location spoofing.
ConnectTVT 50 Gamechangers winner 2020
TechNation Rising Stars winner 2019
Thames Valley Tech Awards winner 2018
In the Press
The Economic Times
AI can now create Art but fact-checkers are worried
MIT Technology Review
Deepfake-busting apps can spot even a single pixel out of place.
Fake-News Makers, Beware: This Firm Claims It Can Oust A Photo With One Fake Pixel
Startup Street: Serelay Claims To Bust Fake News In Real Time
Startup develops software to identify fake news counterfeit photos
New technology combats deepfake videos
big thing: Tracing deepfakes
Wall Street Journal
‘Deepfakes’ Trigger a Race to Fight Manipulated Photos and Videos