Your browser is not supported. Please update it.

21 November 2019

iQiyi goes it alone for watermarking, passes ChinaDRM Lab tests

China was late waking to the challenge of video streaming piracy but has taken up the cudgel over the last few years and now its iQiyi streaming service has claimed to be the country’s first online platform to receive security accreditation from the ChinaDRM Lab for its own in-house watermarking technology. This raised eyebrows because in March 2017 iQiyi announced it had selected Irdeto’s Tracemark technology as the basis of its forensic watermarking to combat redistribution piracy, which has become the scourge of OTT. It now looks as if iQiyi has displaced Irdeto with its own in-house watermarking technology to avoid dependence on US IP (Intellectual Property) in the wake of escalating trade wars between the two countries. Irdeto did not respond in time to our request for clarification on that point.

The move is also significant in the context of China’s own distinct security requirements, which go beyond merely protecting against illicit content redistribution. Internet video piracy has been rejuvenated in two waves, firstly by the SVoD boom and then more recently by proliferation of streaming linear content, especially premium live sports. The latter has led to the emergence of a new arms race around live streams whose value and even existence often only lasts for a few hours. This means pirates can jump around between sites and are not confined to dark parts of the web because many of the big social media platforms have done relatively little to clamp down seriously on the problem of intermittent unauthorized streaming.

In China, the authorities are able to control and even manipulate news behind the country’s “Great Firewall” to an extent unprecedented in any other major power. That has been evident during the ongoing Hong Kong protests where a lot of disinformation emanating from the Chinese government has circulated internally, the main point being that foreign countries, especially the UK, have been accused of stoking up the protests.

Another distinctive feature of China is that when the authorities do decide to act, they move fast. That was the case for video piracy when it became clear the phenomenon was causing economic damage to home businesses and not just foreign studious or content producers. That became clear not just for streaming or even linear pay TV but for the even more venerable business of the cinema.

For years the Chinese box office had been gaining steam as the country’s expanding middle class flocked to see both Hollywood-produced and home-grown movies in a country where attendances at traditional theaters has always been high. Annual box office receipts had been growing at over 30% for several years until 2015 on the back of strong economic growth and an associated boom in both content production and importation. But this spawned streaming piracy with the blockbuster titles becoming readily available online and this was the main factor in an abrupt slowdown in box office revenue growth in 2016 to almost zero.

Then the authorities sprang into action with a large-scale network forensics operation conducted by the National Copyright Administration of China (NCAC), which first identified 3,000 websites to monitor closely. Of these, 58 major platforms, including some large video and music streaming sites as well as cloud storage providers, were singled out for close attention, leading to closure of 700,000 web links during 2017. Cinema box office sales rebounded, growing at 13.45% in 2017 and 9% in 2018, admittedly not back to earlier levels but then from a higher more saturated base coupled with a drag imposed by availability of movies within shorter windows from legitimate streaming sites.

A notable underlying trend was the swing towards China’s domestic film economy, which accounted for 54% of all box office revenues in 2017, rising to 62% in 2018, according to the country’s State Film Administration, again stoking up demand for effective anti-piracy.

The NCAC and other agencies have also come under growing pressure to combat sports stream piracy both from foreign leagues and increasingly some indigenous leagues as well. This reflects rapidly growing consumption of sports and associated rights values, as our research arm Rethink TV highlighted in the recent report Globalization lifts TV sports rights past $85 billion futureSports Rights Forecast to 2025. This forecast that Chinese media rights for English Premier League (EPL) soccer would balloon from $350 million in 2018 to $931 million in 2025 and that US National Basketball Association (NBA) rights values would quadruple from $90 million to $362 million over the same period. Local Chinese Super League soccer, despite problems attracting top foreign players during their prime, is also set for significant growth in media rights values from $260 million in 2018 to $445 million in 2025.

As in other markets, these revenues are jeopardized by stream piracy, which has surged with the proliferation of sports streaming. It is certainly true that in the absence of hard data on piracy, which like the black economy is by its nature impossible to count accurately, some of the wilder predictions are exaggerated. They come from content security firms, leagues themselves and others with a vested interest in hyping the issue, but even so there is clear evidence that content redistribution piracy has increased significantly over the last few years.

Therefore, having clamped down quite successfully on VoD piracy, China’s service providers such as iQiyi have ramped up efforts to do the same for illicit live stream redistribution. This is where the forensic watermarking comes in and explains why iQiyi as the country’s largest sports streamer is eager to trumpet its advances, for forensic watermarking is as much a deterrent as a direct defense. Having launched its first paid content service in 2011, iQiyi had gained 20 million subscribers by 2016, mushrooming to become the country’s first streamer passing the 100 million mark by June 2019.

Its own internally developed watermarking system has passed several tests quite similar to those required for certification by MovieLabs, including successful embedding of a 64-bit arbitrary watermark into video clips and being able to retrieve those marks from random parts of the frames. It also had to demonstrate embedding of marks without discernable visual artifacts. Furthermore, it has had to demonstrate resistance to attacks on the watermarking itself, which is where the latest arms race is taking place, involving techniques that have evolved significantly since MovieLabs first, in 2014, mandated forensic watermarking for rights holders wanting to distribute its members’ content at 4K quality. MovieLabs has now endorsed the security assessments of ChinaDRM Lab and therefore iQiyi’s watermarking implementation.

It is notable how Chinese research groups have become significant producers of innovations in watermarking as reflected in the technical literature. We have not been able to define exactly whether iQiyi’s watermarking technology has benefited from such work, but it may well incorporate some developments from a seminal open access 2017 paper authored by researchers from three Chinese academic groups including the Qufu Normal University in Rizhao, Shandong Province (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0196306).

This focused on optimizing the balance between the two conflicting demands of a watermarking system, robustness and invisibility. The latter is clearly important to avoid affecting the viewer’s experience and is measured by PSNR (Peak Signal Noise Ratio) where the watermark is considered as noise disrupting the overlying video signal. Robustness reflects ability of the watermark system to remain fully functional upon transformation of the video, which could occur during processing such as reducing resolution, by accidental loss of data during transmission, or by deliberate attack. Generally, robustness is proportional to the degree of transformation and therefore makes it harder to avoid being perceived by the viewer.

As a rule of thumb, modifications to an image are invisible to the human eye when the PSNR is greater than 30. The Chinese paper describes a method under the general heading of differential evolution to optimize the balance between robustness and invisibility such that both meet their targets, including the PSNR level. It is an iterative process inspired by biological evolution to improve fitness by obtaining a combination of variables that meet broad requirements. In this case there are two variables whose value is determined by the extent of the watermarking transformation. The algorithm continually tweaks that watermarking process until values of both variables meet set targets. The algorithm is not good at determining the theoretical best or most optimal solution, but as in biological life the goal is to achieve something that is adequate for survival.