TOKYO, August 31, 2023—Thomson Reuters, a global content and technology company, Canon Inc., and Canon Europe Ltd., a world-leading provider of imaging and information technology solutions, and Starling Lab, an academic research lab based at Stanford and USC innovating with the latest cryptographic methods and decentralized web protocols, announced today the completion of a pilot program demonstrating how news organizations could certify the authenticity of an image and ease concerns about content's legitimacy.
In collaboration with Canon, global news organization Reuters proved its ability to embed trust into a photograph from its source by preserving a picture’s data provenance chain of integrity – the image’s metadata – from the point of capture to publication. Reuters also successfully integrated Starling Lab’s authentication framework into its picture desk workflow. One of the first practical news-gathering applications of the technology, this pilot comes at a time when consumers are increasingly worried about their ability to distinguish between real and fake news on the internet, and recent advancements in generative AI technologies which are making it easier for anyone to create visuals to deceive or misinform viewers.
“Trust in news is critical. However, recent technological advancements in image generation and manipulation are causing more people to question what is real, and what is not. Reuters continues to explore new technologies to guarantee that the content we deliver to the world be factual and trusted,” said Rickey Rogers, Global Editor, Reuters Pictures. “Our collaboration with Canon demonstrates the potential for new technology in image verification to increase consumers’ confidence that what they see is genuine. We look forward to sharing the encouraging findings from the project with our customers and applying them to protect our photojournalism. Our goal is to remain one of the most trusted global news sources.”
The pilot project saw Reuters photojournalist Violeta Santos Moura capture pictures using a prototype Canon camera that digitally assigns each photograph and its corresponding time, date and location with a unique identifier (hash value) and then cryptographically signs them to establish a root of trust for their authenticity. The photos are then registered into a public blockchain and updated after each modification by the Reuters pictures desk. This process continues until the photo is distributed with its metadata, edit history and blockchain registration embedded in the photograph using the new C2PA standard. To verify the picture’s authenticity, news customers can compare its unique identifier (hash value) on the public ledger.
“Many photojournalists rely on Canon technologies. We understand the role that images play in society and recognize the importance of preserving image authenticity”, said Richard Shepherd, Product Marketing and Marketing Strategy Senior Manager, Canon Europe Ltd. “Working as part of the Content Authenticity Initiative (CAI), Canon wants to take meaningful measures to protect image outputs that serve the news community worldwide.”
Source: Canon Global
This application makes a lot of sense.
It is not like they are making Bored Apes.
Moreover, film days you can mask and collage on an enlarger anyway (analog photoshop), just that it was much harder and time consuming so it wasn't much of a problem.
I do like the idea of blockchain authentication of every step of process. It would be even better if we can use it to trace back and see the original captured source image prior to edits, which in turn may have news organizations streamline their image workflow and just show the base capture, perhaps with some crops and call it a day.
The worry is though that the number of images captured nowadays are staggeringly high and handling the data volume may be an issue.
It's one thing to embed a chip to capture that information and sign it with a trusted cert chain. (Perhaps using the Helium network to publish the hash in advance of the release of the image itself and to provide verification of date/time/location of capture as well?)
But it's an entirely different problem to have that then follow the image as it's edited from RAW, compressed, shared, etc.
The level of trust required that the camera's firmware wasn't modified or fed a fake image...
I'm certain they are thinking about all of this and more.
The end goal of non-repudiation for photos is incredibly challenging in today's world.
Great article and an awesome direction to see movement towards - it's only scratching the surface though and the techie in me wants to see more details.
Politicians are already one of the least trustworthy professions (ipsos veracity index) so elections will be a prime target for malicious ai. I predict the rate of votering (in countries where it is optional) will decrease further. It will be interesting to see what happens where voting is compulsory
RAW image files were intended to be used that way.
I guess we still could do that if we wanted to.
They can easily be copied so we would just need to secure them somewhere.
The other thing that we could do is keep the CR3 files and share DNG files when clients request RAW files.
Programmers use it all of the time.
It just got into the public eye with NFT.
NFT is also useful but trading NFTs or images without any actual ownership of the images was just plain silly.
AI has been around for decades as well.
GPT is pretty new and now AI is the new thing.
A bit like ‘cloud computing’ being ‘new’
Now, if Reuters and Canon can get AP, Magnum, Sony, Nikon, NYT, Le Figaro, CNN, BBC etc. on the same boat, and build a blockchain where any change to an image is recorded into a distributed system not controlled by a single entity, a blockchain could make sense. A blockchain run by Reuters itself only doesn't make any sense. Anyway, one of the problems of a real distributed blockhain is the time it requires to approve a transaction. So, it is feasible probably to record everytime an image is "published" to the blockchain. Blockchains also don't play well with large data, so storing whole RAW/TIFF images into a blockchain won't work very well. C2PA is designed to store a thumbnail of the images inside is data - it is enough to identify all image changes properly?
I'll add auto license plate blurring on speed trap cameras if you pay your local politician enough to the list!
/s