The role of AI in content protection

The threat of illegal content redistribution really cannot be underestimated. That’s the consensus from yesterday’s IBC Conference technical paper session on cyber and content security.

So let’s look at some facts – according to Irdeto data from late last year, there were more than 2.7 million advertisements on e-commerce websites, including Amazon, eBay and Alibaba for illicit content streaming devices. This, combined with the exponential rise in the number of illegal streams from peak live events where traffic multiples in a short space of time, creates a huge challenge for content owners and rights holders. As the threat rapidly evolves and grows, security strategies need to evolve in tandem.

According to Denis Onuocha, Chief Information Security Officer at Arqiva, in the cyber and content security session: “The notion of the perfect defense doesn’t exist, you need to know how to detect and respond. It’s a risk-based approach.”

There are multiple technologies and proactive services that need to be implemented, and one of the key steps is effective monitoring. “We look at the integrity and quality of streams yet nobody looks at why this encoder is talking to that server,” Onucocha said. “We need to spot these anomalies and then work backwards.”

One proven method of detecting an illegal stream is to send out web crawlers to find any and all streams of a particular event and then to determine – by identifying the broadcaster logo and tracking validation back to source – whether the stream is pirated. If so, take-down notices can be issued to the offending ISP.

However, the scale of live event broadcasts and illegal distribution is now so great that doing this manually is next to impossible. In the same IBC Conference session, Irdeto CTO Andrew Wajs outlined how Artificial Intelligence can play a key role in providing scale.

There is a further problem which needs to be tackled though. The encoding and processing employed by pirates (or those simply using a camcorder) creates artefacts and often blurs the logo rendering recognition tricky. Therefore, the AI needs to be trained on just the right data set to be able to overcome this. Irdeto has trained its algorithm on two of the main industry standard neural networks (NN) known as AlexNet and ResNet as well as on its own NN, IrdetoNet.

“It is really important to train the AI across as many logos as possible,” explains Andrew Wajs, CTO, Irdeto. “The results just based on IrdetoNet, deliver an accuracy over 98% and because it is trained on a refined media specific data set it needs less computer power and therefore less cost.”

Irdeto says it combines this innovation with its proactive Cybersecurity Services to reach a new level in the fight against piracy.