Is Blockchain a solution to image rights management? And if the Webcrawler was KODAKOne’s key idea, would not Image Recognition be its weakest link?


At CES this year, KODAK announced the launch of the platform KODAKOne managing copyright of images based on the blockchain technology. Photographers can rent a machine, so-called KODAK Kashminer, to be part of the platform. Some find this initiative extremely promising, others treat this announce as a scam.

 

Let us quote the press release:

“The KODAKOne platform will create an encrypted, digital ledger of rights ownership for photographers to register both new and archive work that they can then license within the platform. With KODAKCoin, participating photographers are invited to take part in a new economy for photography, receive payment for licensing their work immediately upon sale, and for both professional and amateur photographers, sell their work confidently on a secure blockchain platform. KODAKOne platform provides continual web crawling in order to monitor and protect the IP of the images registered in the KODAKOne system. Where unlicensed usage of images is detected, the KODAKOne platform can efficiently manage the post-licensing process in order to reward photographers.”

 

We will not discuss here the commercial move of KODAK (this initiative has nothing to do with the former well-known KODAK company as it has sold its branding to other businesses). We will not discuss the economical aspect of the deal (renting a KODAK KashMiner machine is certainly not as lucrative as promised).


 

Can blockchain technology be a solution to image rights management ?

The KODAKOne platform is indeed offering the opportunity for photographers to be part of a system based on 3 pillars:

legal-hammer A blockchain relying on a network of KashMiner machines. The blockchain stores the Intellectual Property (or IP) transactions of the images registered in KODAKOne.
corthouse A cryptocurrency called KODAKCoin. All transactions are valued in KODAKCoin.     This creates a new economy centered around image rights.
 web-search

A webcrawler scouting unlicensed usage of registered images.

 

Blockchain is the new hype… but what is it?

Today, photographers rely on photo agencies to manage image rights: to trade usage of images against money, to record these transactions, and to give the money back to the photographers. Trust in the photo agency is key, and the photographer has to pay this trusted third party.

KodakOne (so as Binded.com) aim at replacing the photo agencies by a blockchain. A blockchain is nothing more than a register storing transactions, ie. a ledger.

 

ledgerJohn Carlin [CC0], via Wikimedia Commons

 

This ledger is not kept by one trusted entity, but it is decentralized over a network. Anyone on the network can read the ledger, but no one can modify a transaction once recorded in the ledger. See this video for a technical introduction to Bitcoin and the blockchain technology.

In a blockchain, the nodes of the network interact according to a complex cryptographic protocol. There are two main drawbacks:

 

 speed This protocol is slow. Few transactions are registered per second. In the Bitcoin system, this rate is around 7 transactions per second. This is ridiculously small compared to VISA scaling as millions per second.
 energy This protocol is consuming a lot of power. The bitcoin network continuously drains more than 3GW. Over one year, it consumes more electricity than a country like Ireland.

 

These are the prices to be paid for having a secure decentralized ledger.

 

A blockchain does not run for free.

Being an active node in a blockchain network consumes a lot of time and energy. This deserves a big reward. Nodes receive incentives when they succeed to register transactions (the operation of “mining” a new block). For instance, in Bitcoin

  • They collect fees charging the parties involved in the transaction.
  • They create new money that they keep.

This is certainly the reason why KODAK introduces a new cryptocurrency, KODAKCoin.

 

In the end, photographers will have to pay. It is not very clear whether IP management will be cheaper with a blockchain than with a trusted photo agency. 

It is not very clear as well why photographers absolutely need a decentralized system…

 

So, is the WEBCRAWLER the key idea behind KODAKOne?

The main advantage of the blockchain is that the ledger is public. Therefore, anyone could theoretically check whether a particular usage of image is duly licensed by KODAKOne.

This opens the door to a decentralized webcrawling service. IP monitoring over the web could be done in parallel by many nodes. This will be a very powerful tool if some technological bottlenecks are fixed:

  1. Nodes should share which part of the web they crawl, or there should be an entity managing the crawlers to efficiently cover the web.
  2. Crawling the web consumes computational resources. What will be the reward?
  3. How does the webcrawler recognize an image found on the web? How does it learn that this is a copy of an image registered in KODAKOne?
  4. When image is recognized, how does it check that the usage is licensed? What is the content of the transaction? How is the buyer identified?

 

The pressure is on image recognition

At IMATAG, we know how tricky point #3 can be. Indeed images found on the web are never an exact copy of their registered version due to some cropping (to remove a visible watermark for example), resizing, and compressing (to speed up web page download), mirroring, color enhancement …

 

These kinds of image processing are very common, and the robustness of the image recognition is paramount to tackle the following three issues:

  • Wrong identification: Image recognition has difficulty distinguishing too similar images. Imagine two photographers side by side taking a picture at the same moment of the same scene. Imagine a photographer registering a picture in KODAKOne, but also selling it via a photo agency. Image recognition will not distinguish the two versions.
  • False positive: the system is wrong when stating that this image on the web is a copy of a registered picture. The artist, Sebastian Tomczak, has recently uploaded a video on youtube which CONTENT-ID (the image recognition system run by Google) deemed as copyright infringing. The problem is that the uploaded video is indeed pure noise.
  • False negative: the system doesn’t recognize that image on the web as a copy of this registered image. The case of Eduardo Martins deluding Google image search by simply mirroring image is a typical illustration.

 

The press release says that WENN Digital, the other partner behind KODAKOne, is experienced in AI-enabled image recognition. Be very careful about AI. Many research works recently showed that simple dedicated attacks easily delude image recognition based on AI. This is the hot topic in research right now. Simply type “Adversarial attacks” on google.

 

For all these reasons, we think that the image recognition, bridging the images found on the web and transactions recorded in the blockchain, is the weakest link of the system.

 

We use invisible watermarking because of its robustness (test it here) and security. Though this technology is centralized by design, it is not incompatible with a blockchain service, as proven by this initiative in the music industry : Dot Blockchain Media Makes Blockchain Plus Watermarking a Reality.

 

Conclusion

If the reason to choose the blockchain technology for image rights management is decentralisation, it has to be fully justified against the consequences: the cost for each miner implies a complex rewarding system. KODAKOne made the choice of a business model based on an ad hoc (thus questionable) currency. Other startups also decided to leverage on blockchain’s promise (or hype), will each of them create their own currency?

Considering that images illegally used on the web are usually modified, which technology will efficiently identify copies of registered images? Image recognition based on AI and fingerprinting might be too weak, that is typically a case where invisible watermarking is relevant.


Also published on Medium.