In the post-epidemic era, education has turned to online, short video and TV drama market, and the attention paid by audio and video platforms to content protection is unprecedented. In China, with the support of 5G, video consumption is booming and users’ payment habits are forming. How to increase and protect revenue has become the focus of attention of various video platforms and online education platforms. This article is from the speech Shi Guang, product director of Aidid Asia Pacific, shared on LivevideoStack online. Starting from different video content piracy methods, he discussed with everyone how to achieve end-to-end protection of video content creation, distribution and consumption.

The text/applied widely

Organizing/LiveVideoStack

I am very glad to meet you in this forum at such a time. My topic today is the best practice of global multimedia video content protection. Aidid is a company headquartered in Amsterdam, the Netherlands, so we have some global practices to share with you here.

01

–

The development of the streaming media market

Above is issued, Internet streaming media market (left) is a global growth over the past three years, is expected in 2021 $180 billion + the size of the market, and China in 2021 were estimated at $25 billion + the size of the market, and the right of the Chinese market is more advertising revenue than user subscription revenue, it is an important characteristic of the Chinese market.

The epidemic has been a huge boost to our industry. The total volume of the industry will grow rapidly from 2020 to 2021. The most important driver of the industry’s growth, in the final analysis, is a wide variety of high-quality original content. But the openness of the Internet means that streaming content, if left unprotected, can become virally pirated and redistributed. So it’s important to protect high quality content. In recent years, China has paid more and more attention to the protection of content, especially in the first half of this year, there were some interesting topics of copyright protection.

This is some from the Internet to find the topic of copyright protection, there are various views on the copyright of long video, video, short video, the specific details will not say more. In fact, for the past three to five years, the battle between the long-form video platforms has been very intense. Except for Mango’s earnings, most platforms are still losing money. In the past two years, short video has been a new entry into the battle, and the form is even more complicated. In any case, the long video platform of copyright protection understanding, awareness, attention and investment are relatively high, we also see it. Short video platforms have also gradually realized the importance of the need for copyright protection, including videos like Li Ziqi’s, as well as some videos on official accounts and B websites, which have also been removed. Competition between long and short videos is inevitable, but the picture above shows that copyright protection is very important and needs long-term planning. No matter it is long video, short video or online education platform, in fact, there are common demands and pain points for content protection.

There is no doubt that the first is to protect the revenue of the platform. If the content with high value-added is easily cracked and spread by piracy, the revenue will drop dramatically and the income-output ratio will be extremely mismatched. And foreign or domestic carriers in the process, we are aware of a problem, copyright owners distribute its content to the platform operator, copyright owners claim to the copyright protection is actually very demanding, both movies, sports, games, competitions, etc., in this case, like we irdeto content protection supplies that can provide a whole package. It doesn’t seem to be particularly pleasant to the carrier because it requires a certain amount of integration work, but there’s actually a lot of value to it. So preventing illegal downloading of content, preventing screen recording, tracing the source of piracy and so on are the content protection appeals that have reached consensus.

I want to share with you some of our best practices. The picture above from left to right is an end-to-end chain of copyright protection. Content watermarked and protected by DRM is likely to be pirated by pirates after it is disseminated through the carrier platform. And what can we do when our content is distributed by pirates? We can do a whole range of things from prevention, detection, detection, to taking them off the shelves. For example, the first step is the cloud crawler, which can search the pirated content on the Internet 24 hours a day. This is the process of discovery. Discovered need to analyze, there are three levels, the first is to extract the fingerprints of the video content, the second is the watermark detection, according to the results of the watermark detection can detect the pirated video streaming from the only identifies the pirates to forensic watermarking, sometimes there are also some artificial expert analysis, this is to verify the process of c data, The data collected can be used as evidence for legal proceedings, so it is necessary to collect these data. The next step is to get through with law enforcement, mainly focusing on how to quickly find the source of piracy, send the corresponding notice of removal, and communicate with the police and legal departments so that pirates can stop piracy and comply with the law as soon as possible. And finally, business intelligence. We have just sorted out not only technical solutions and means, but also relevant processes such as forensics and law enforcement. This shows that we have an understanding of end-to-end copyright protection based on practical experience.

02

–

DRM digital rights protection

Let’s take a look at DRM. DRM stands for digital rights protection, which in its simplest sense means that video streaming is encrypted. If the Content is encrypted, the question that immediately comes to your mind is, how can the Content Key that encrypts the video stream be transmitted? The simplest form of the Key to encrypt the video stream can be passed down in clear text, and carried by another corresponding Request and response message. This is also the simplest Clear Key Request method, which is obviously not very secure. In order to make the secret key of encrypted video stream safe transmission down, it is necessary to terminal server side on authentication, mutual trust and encryption and decryption communication interaction. You need to distribute the content key safely. That’s the basic idea. No matter you are in what platforms, even if is the monthly pay, also can have the latest blockbuster to pay on demand, on demand after the success of the pay system will tell you to watch video time effectively, and only after playing for the first time in 24 hours to see all the and so on, all this is content distribution strategy, which is also the category of DRM. In addition to the content key encrypting the transmission, a number of distribution policies are also encrypted.

Back to the image above, Apple has the HLS protocol, and the corresponding DRM is called Fairplay Streaming. FairPlay has been protected for music copyright since the Jobs era, around 1990. But in 2015, Apple launched Fairplay Streaming DRM technology, the name of which actually emphasizes copyright protection under adaptive Streaming media. Its native DRM technology and its operating system and platform are very well integrated, all developed by Apple itself, the entire ecological chain is controlled by Apple.

The whole frame on the right is Dash, and the yellow box is WideVine. WideVine is a company acquired by Google in 2010, which was specializing in DRM solutions at that time. After Google’s acquisition, WideVine was integrated into Android operating system 4.4 or above, which is relatively perfect and stable. In addition to the physical devices in the yellow box, Chrome, Opera, Firefox and the new version of Edge are the browsers framed with a dotted line in the upper left corner. These browsers, too, have built-in WideVine, a software solution, of course.

On the right, Microsoft is in a similar situation, with certain devices that only support Microsoft’s PlayReady DRM, such as the latest Xbox or PlayStation models. We also participated in the China DRM Forum, which is the DRM specification independently defined by China. It has been upgraded to version 2.0 in 2019. It is public information that Huawei, which represents equipment manufacturers, is the biggest supporter of ChinaDRM. Huawei’s Harmonyos and Android operating systems are included, and ChinaDRM is already supported on tens of millions of mobile devices, and it’s Hard Profile. As we have just said, Soft Profile is usually supported in this browser. Hardware Profile is supported in the case of high security chip or TEE chip, etc. It is very complicated, but it is more secure.

For the current situation, the domestic ranking of the top several long video platform, may have been self-research ChinaDRM system and program, because the specification is open. However, for a relatively small volume, the self-research ability is not so complete, including the new online education platform. If you want to study new technologies, how to cover all terminal equipment is very complicated. Companies like ours can provide better service and value by helping operators adapt various DRM technologies to the entire video distribution process, from stream generation distribution all the way to end-side smooth playback. We have a good cooperative relationship with Apple, Google and Microsoft. We can help operators centralize and unify the server side corresponding to DRM technology, and then based on the strategy for content playback, we can also quickly provide centralized and flexible unified configuration and management.

As far as DRM technology and solutions are concerned, what problems does DRM solve, such as illegal downloading? But streams protected by browser-based DRM can still be copied to screens, which are screen copies, because this is DRM technology based on software solutions, and there is no way to fight screen copies. There are many software programs that can copy the screen at any time, especially on Windows platforms. So at this time, what technology is used to deal with it, that is, we will talk about the watermarking technology.

03

–

Video and audio watermarking technology and its application

First of all, let’s clarify that the watermark we are talking about here is an invisible and secret watermark, which can be used for forensics. Before I go into the details of this slide, I want to explain how it works. Watermarking actually uses some transformation theories in information theory. Usually, time-domain signals are converted into frequency-domain signals and then processed, such as discrete cosine transform and wavelet transform. In short, some information, the binary bit string, that identifies a particular individual, is added and inserted into the video stream. The visual perception of the human eye is almost invisible, and the solution’s own monitoring platform can detect it, that is, can detect WMID.

To distinguish from the watermark insertion way, there are two big project, the first is the direct method, from left to right on the drawing there are five pieces, finished with a watermark generator, have more than one specific client, each with a unique identifiable watermark identification code, the superposition of the watermark identification code directly to a video frame or on fragments, No doubt can continue to overlay on different pieces. You may also have some new problems. Should you insert the watermark before the video is distributed or when it is played on the device terminal? Actually, both schemes can be done.

The next step is a two-step process. The first step is the watermark A/B embedder. In other words, the original fragment is added with some A information, and then some B information. A and B are just an identifier, which is equivalent to two copies of the video fragment being copied with A information and B information added respectively. The second step is that when the terminal side requests A video clip, the watermark segment switcher will select whether to output the A side or B side from the dimension of time according to A unique characteristic value of the terminal side, such as the temporary generation of A binary bit string WMID. For example, if the first terminal consumer sends A video playing request and the watermark ID is 00110, then the service side will select the information A and B, and select the corresponding A/B according to the chronological order and watermark ID, and so on. In other words, the video stream requested by each user, from the time dimension, is already A personalized and uniquely identifiable video stream, so this is the basic introduction of A/B fragment method.

Because watermarks are invisible and transparent, in addition to the robustness we will talk about in a moment, there are also some subjective and objective considerations about the impact on vision. Foreign countries have this kind of special detection, visual sensitivity beyond the ordinary detection, can see in the end your video has changed, there are some objective criteria, such as peak signal-to-noise ratio and so on. But the trickier bit is robustness, because there are all kinds of attacks, whether your attack is malicious or non-malicious. For example, in the simplest case, compression, noise, transcoding, these are all things that happen during normal video processing operations. These operations may cause loss and attenuation of the added watermark information, resulting in undetectable during detection. Moderate attacks are more difficult. The last thing that needs to be emphasized is malicious attacks or collusive attacks. Take frame interleaving attacks as an example. If you have a group of pirates, they pirate the same film together, take the video, and then re-slice and reassemble it. For example, the first second uses the clip of the pirate X, and the second second uses the clip of the pirate Y. In short, suppose there are five conspirators who randomly combine their video frames according to the time series. At this time, the newly generated video stream is submitted to the original detection platform, which is very difficult to detect. Especially as the number of the co-conspirators grows, the difficulty of detection increases sharply. Of course, this is a theoretical concept, and we do have this kind of attack in practice around the world, but the number of conspirators is small. An overlay attack is similar, in which several conspirators’ videos are overlapped synchronously over time and averaged to produce a new stream of video. Mosaic attack is to divide the videos of the nine possible conspirators into nine grids. In the first grid, small video blocks of A are placed, and so on. A video stream is randomly reassembled, and the watermark information may not be detected. In short, the principle of watermarking scheme sounds relatively simple, but it is very complex to do a good job. At this point, manufacturers in the industry have their own schemes.

Just now we have talked about the most basic principles, and then we will introduce the concrete implementation of the practical application of the scheme. The image on the left is the distributor watermark, which is the watermark that is added to the distributor during the post-production and distribution stages of the video. For example, after a Hollywood blockbuster is produced, the film distributor goes to different platforms, such as iQIYI, Tencent and Youku, when it is released, these operating platforms are distributors of the film distributor. Our scheme can add a unique distributor watermark to the blockbusters, so as to detect from which platform the leaked or pirated video stream is leaked in the later stage. Who is the target customer of this project? It is to sell to film and television producers, film studios and distribution companies, such as Hollywood film makers, Alibaba and Tencent, who all have film companies, and their film companies need this kind of scheme. The solution for platform and online education operators is the OTT watermarking solution on the right, while Broadcast solution can be almost ignored in China. The scheme, which is aimed at streaming service operators and pay-TV operators, can identify among a large number of consumers that a particular subscriber is illegally pirating content. The left is to add distributor watermark to blockbusters produced by Hollywood. For example, the recent “Fast and Furious 9” will be released soon. The total number of global distributors is about 100, so the added watermark is measured in hundreds. For the scheme on the right, for an operator platform, such as iQIYI and Tencent, the payment for a single platform exceeds 100 million, so the added water print volume should be measured in billions. These are two completely different schemes.

The distributor watermark is borrowed from a cloud service that uses IBM’s Aspera, a high-level protocol that is many times faster than FTP or HTTP traffic. In general, we have a cloud service, and when different distributors request to download a movie, the cloud service can add the watermark according to the global policy or the personalized configuration, and the watermark can be immediately distributed. In fact, the copyright or distribution company to distribute the content is also at the last moment to release the source, the last moment to leave the distributor time is also very short, there is bound to be a fast download process. Therefore, we cooperated with IBM Aspera Cloud to create such a solution to meet the demand of fast and secure download. The advantages of the solution are obvious: rapid deployment, ease of use, and no impact on user experience.

OTT solution just also said that to some, it can be applied to UHD on-demand content or Live, could do it, “user content controlled comprehensively, and also has good robustness, just for distributor solution but we are not considering the collusion attack, because the amount of the watermark is a total of 100 or so, 100 distributors to collusion is unlikely. However, such an OTT solution must have the aforementioned malicious attack prevention considerations.

04

–

Best practice case analysis

The following is a user case sharing. This is the full protection of a live event that we helped an OTT operator to do on November 9th, 2019. Logan Paul and KSI are actually not professional boxers, but super web celebrity on YouTube, and they each have around 23 million followers. In 2018, they actually fought as amateur boxers. This time, the pro fight will be streamed live on a streaming platform only to subscribers who pay at least $20 a month. In this case, how do we protect the actual scene? The proposal in this case could also be applied to other live matches, including the European Championship, which is currently under way. Let’s look at the details.

This case for live anti-piracy and we speak in front of the best practice content is similar, I will not repeat, only need to emphasize that, in view of the live event, our watermarking scheme has a request, is the ability to rapidly detect pirated video source, otherwise the boxing match, pretty soon, when three games is over, a total lasted an hour, If the detection speed is not fast, it may not be detected after the live broadcast is over, then the scheme will fail. This is a difference between live and VOD watermarking schemes.

For the list that may be pirated, we should do some work in advance, so that we can capture it more specifically in the event. Moreover, we should take out the whitelist of the platform that has sub-authorization and define the interaction rules. In addition, our scheme also connects with YouTube’s content identification system and Facebook’s copyright management, and at the same time deleting the search engine results in advance, greatly raising the threshold for obtaining pirated content.

When live streaming, the solution crawler starts scanning pirate sites, Kodi, social media sites, etc. The content of automatic analysis was also mentioned earlier. When foreign pirates want to reach consumers to promote their pirated streams, social media promotion and publicity are very important channels, so YouTube and Facebook are our focus, just as WeChat is our domestic promotion and publicity channel. Our solution then detects and finds the problem and is able to automatically perform direct removal. For example, CDN, we also have cooperation with CDN, CDN can immediately cut off the pirate stream after the CDN receives our conclusive evidence of piracy and the notice of removal.

For the service after the competition, we need to summarize all the original data, including relevant information, hosting websites and linked websites, including watermark detection, forensics package, search engine downgrade, continuous update, etc., which of course includes the application of DRM scheme we just talked about. DRM is essential in the whole scheme. On top of this DRM, we add video watermarking and a whole set of piracy monitoring and detection system to form a closed loop.

This is the portal of the idid anti-piracy monitoring system, which provides a variety of key information, including compliance, relevant percentages on social media, and the status of the corresponding linked and hosted sites. These can be refreshed in real time, because there are crawlers, our piracy detection system and watermark detection system and other continuous dynamic input, and all the real-time data are presented here.

Now, a quick look at the ROI for the overall solution, which is the actual case data, the KSI vs. Logan-Paul match was viewed by 214,092 people on foreign social media who had their pirated streams taken down. In 24 hours, there were 93,089 searches for “free streaming” in the browser, and those two combined could potentially be 300,000 paid viewers. These 300,000 users are identified as potential users to calculate the ROI of the game at 1%, 3%, and 5% conversion rates, respectively. It can be seen that the input-output ratio is quite high, and it is very important and valuable to use this system from the perspective of operators.

Last but not least, some key business data from Idid. Over the past year, we’ve secured 22.3 billion OTT video streams. We actually offer cloud hosting services on a variety of DRM servers across the globe, on every continent. Chinadrm also has a little bit of it. We protect more than 6 billion devices and applications, and we have core proprietary technology that can protect devices and applications. So far our online crawlers have detected 4.5 billion violations, all of the data is currently outside the country, and we have issued 280 million notifications on behalf of copyright holders.

At present, Aidid mainly serves two large markets. On the left is the video entertainment industry. Relevant data are not repeated above. The new business direction is security protection for the Internet of Things, including the Internet of Vehicles, etc. Aidid is a 52-year company headquartered in the Netherlands, with offices in 14 cities around the world and an office in Beijing, which is the central node in the Asia Pacific region.

That’s all for this time, thank you.

Q&A:

Q: Compared with digital watermarking, will the amount of data needed to be calculated and processed for video fingerprint generation be particularly large?

Mike: It’s possible. We don’t do fingerprints, we work with third-party vendors. We believe that this part is not the most core compared with the highly complex encryption and decryption technology, so we choose to cooperate with the third party manufacturers. Of course, my understanding is not necessarily accurate, I think the watermark processing method will be more computations than fingerprint.

Q: Does watermarking of so many videos not require encoding and decoding? Will the performance hold up?

Mike: For OTT watermarking, there are two schemes: server-side watermarking and terminal-side watermarking. Terminal side watermarking is not discussed in detail. This scheme refers to adding watermarking information in the part of Media player, but this must be done in a secure environment (such as TEE). On the server side, the watermark A/B snippet switch will usually select key frames instead of all frames. For example, usually A film with A/B watermark in the final size of the CDN will be 3% to 5% more. Here is the need to balance, to seek a dynamic balance: if not handled properly, the detection speed is not fast enough; Watermark information added more although easy to detect, but CDN storage also takes up more space, are the cost.

As far as performance is concerned, it is not feasible to deal with malicious or collusive attacks in live streaming. The network transmission of high-value movies of VOD is destructive. For VOD, more watermarks are added in advance on the server side. In A/B selection, the requirements for terminal consumers and CDN performance are not high.

Q: Can the piracy of optical rebroadcast be solved?

Mike: It’s called a remake. It’s a problem that has to be solved. We put remakes in the category of simple attacks (not necessarily accurate), and by simple we mean things that can happen in the course of everyday operations, so they can be resolved.

Q: DRM data is protected, how to steal and forward? No video, no screenshots. If you use another device to record, then another device to record, is the watermark still there?

Mike: DRM data/video streams are protected, and pure hardware solutions make it difficult to steal and forward screenshots. But the pure software DRM scheme refers to the browser playback. There are a lot of screen recording software on Windows. Because of the good openness of Windows platform, these screen recording software can record screen very well. If you use another device to record and then use another device to record again, there will be some problems. In practice, if the definition of the recording can be guaranteed, that is to say, the transcription device is still and stable to record, then the detection success rate is quite successful. But if it is held by hand, as mentioned above, A/B transform, add to the key frame, hand shake A little may have A lot of harm to the watermark. However, hand shaking on the video shooting effect is not good, the quality is also lost, the value of natural piracy value greatly reduced, this is also a balance.

Please scan the picture __ for detailsQr code__ or click __Read the originalLearn more about the conference.