Abstract: From April 11 to 12, 2018, the 2018 Asia-Pacific CDN Summit was held in Beijing. The conference was composed of seven major parts, including asia-pacific CDN Leaders Forum, TV Cloud Forum, Short Video Forum, Video Cloud Forum, New Technology Forum, Operator Forum and International Cloud Forum. In the video cloud forum, Ali Cloud video cloud senior algorithm expert Huang Haiyu made a speech entitled “Narrowband HD 2.0 – let live more amazing magic” theme.
On April 11-12, 2018, the 2018 Asia-Pacific CDN Summit was held in Beijing. The conference was composed of seven major parts, including asia-pacific CDN Leaders Forum, TV Cloud Forum, Short Video Forum, Video Cloud Forum, New Technology Forum, Operator Forum and International Cloud Forum. In the video cloud forum, Ali Cloud video cloud senior algorithm expert Huang Haiyu made a speech entitled “Narrowband HD 2.0 – let live more amazing magic” theme. The following is the speech:
I am Huang Haiyu from Aliyun Video Cloud. My current work is mainly related to aliyun video live broadcasting technology. With the development of live broadcasting, the bandwidth of CDN, the cost of CDN and the definition of video have become a very important issue. The title of my speech is “Narrowband HD 2.0 — The Magic to Make Live broadcasting More Amazing”.
In fact, Ali Cloud has proposed the concept of narrowband HD as early as 2015. After the concept was introduced, many companies started to adopt the concept and launch similar services.
So why did we come up with narrowband HD? In the past two years, what progress has been made on this basis? I’m going to share it with you today.
When you watch a video, you want the picture to be clear, but actually, when you watch a live Internet broadcast, what you see is more blurry video. So, is there no way we can make hd video right now? The answer is no.
Last month, Alibaba Cloud held its first 8K live broadcast at the Shenzhen Cloud Conference. The screen on the stage displayed live images and videos transmitted from Alibaba’s Hangzhou campus, 1,300 kilometers away, and the audience could clearly see the exquisite picture of Cloisonne Taobao figures. However, in actual live broadcast business scenarios, due to network transmission reasons, every audience cannot see a very clear video during Internet live broadcast. The limitation of high-definition live broadcast mainly depends on network bandwidth.
Data from a live broadcast platform built on Aliyun shows that only 16 percent of users can watch ultra hd videos, 54 percent can watch HD videos, and 30 percent can’t see HD videos.
If we can move the ultra hd bit rate closer to HD, will more people be able to watch ultra HD video?
In fact, we’re working on that, but it raises the question of how much of a difference does it make when you move the ultra hd bit rate to HD? Video codec friends will be familiar with the PSNR formula, which is widely used to measure the coding quality of different encoders. In fact, this formula really objectively describes the loss of information after a video is encoded, which is analyzed from the perspective of information theory. However, the actual video is seen by human eyes, and the sensitivity of human eyes to video is not directly corresponding to the amount of information lost in the process of video coding.
Narrowband HD – video coding technology based on the subjective perception of optimal human eyes
Based on the above discussion, the concept of narrowband HD can be derived. Narrowband HD is actually a set of video coding technology based on the optimal subjective perception of human eyes, which studies how to pursue the best visual perception under the condition of limited bandwidth.
When you narrow it down to the subjective perception of the human eye, and bandwidth constraints, you find a lot of interesting things.
First of all, the development trend of coding standards is the same as that of today, and the effect can be superimposed. Encoder development time, almost every decade, encoder will develop a generation, coding quality will save about 30%-50% bit rate, this bit rate saving is in accordance with the PSNR evaluation standard. But the use of narrow band HD technology, focus on the feeling of the human eye, the industry can waste bandwidth savings, let our bandwidth more in the human eye interested in the place, can easily reach 30% bandwidth savings, almost equal to the progress of a generation of encoder. Because the technical focus is on the human eye, we can make the complex scenes that are generally not well handled by the industry beautiful, and the non-professional audience can see the advantages at a glance.
Narrowband HD 1.0 – Two sets of human vision models
Narrowband HD 1.0 is a version released by Alibaba in 2016. Hidden behind narrowband HD 1.0 are two human visual models:
The first model is called the world model between fidelity and subjective score value. The abscissa is the video fidelity, representing the difference of information between the encoded product and the original video, and the ordinate is the score of the video by human eyes, which is 5 points in total. As the encoded video is more and more close to the original video, the human eye is more and more high to the requirement of video, but it is not a linear shape, is divided into three sections of curves, among which ali cloud know this period of very steep better rate of local area is set up, a flat at both ends is increasing video bit rate, improve the objective definition video, It doesn’t actually give the human eye very good clarity. Increase the bit rate a little bit in the middle, and the video will look completely different to the human eye, and each bit rate will be very valuable. Therefore, When setting the video bit rate, Ali Cloud will try its best to select the region with the video tap in the middle.
The second model is the resolution and bit rate model. In this model, the concern is how much sharpness should be set for the same bit rate, or what bit rate should be set for a given sharpness. In traditional video websites, the usual setting method, such as 600K for 480p, 1.2m for 720p and 2.5m for ultra hd 1080p, is similar to such a relatively fixed bit rate setting. In fact, each video requires different bit rates, and the definition of expression is different. For example, for an educational video with an uncomplicated motion scene, a 1080p bitrate will work very well. In practice, we expect to choose bitrate and resolution based on the type and sharpness of the video, while making reasonable bitrate allocations based on the content of the video.
With traditional methods, you usually set a resolution of 720p and a bit rate of about 1m. In fact, the video is pretty bad, with psy’s shirt covered in mosaics and the snowflakes all blurred out. The reason is that the video motion is very complex, with only 1m bit rate. In this case, it is necessary to use 1m bit rate to express 1280×720 pixels. The performance of each pixel will be very poor, and many pixel details will be lost, resulting in the overall effect.
For the later effect, only 887K bit rate is used, which is 15%-20% less than the previous bit rate. The resolution of 432P is used. Such resolution is often set in narrowband HD, in pursuit of the best subjective effect felt by human eyes.
Narrowband HD 2.0 — Go deep into the micro world and analyze video details
Narrowband HD 2.0 was released by Ali Cloud at the 2017 Hangzhou Cloud Conference. Narrowband HD 2.0 goes down to the micro level, analyzing every detail of a video and discovering how sensitive it is to the human eye. The concept of human eye sensitivity model is introduced to focus on which details the human eye pays attention to and which details it ignores. Higher bit rates will be given where the human eye is focused, and some fixes will even be made to the original video to make it more suitable for the human eye experience. You can lower the bit rate for areas you ignore.
From our subjective evaluation, it can be found that h.264 based narrowband HD 2.0 has completely surpassed x264 Veryslow effect and is comparable to X265 Veryslow effect. Through narrowband HD 2.0, we can go beyond the current encoder and advance into the next generation encoder.
Let’s take a look at narrowband HD 2.0 in action. In the picture below, in the focus area, the face and hair of the anchor will appear more smooth with the narrow-band HD 2.0 effect. In some defocus areas, narrowband HD 2.0 allocates a relatively small bit rate without affecting the subjective perception of it by the human eye.
Another real scene is the live game. Shoot-out type game broadcast consumes a lot of bit rate, 1080p set 2.9m game broadcast, the overall will have a lot of Mosaic, because this video at 2.9m, completely unable to express 1080p pixels, narrowband HD 2.0 uses 720p resolution, the same 2.9m, the effect is much better, in addition, Narrowband HD 2.0 will be more sensitive to the human eye text, texture enhancement, the whole video looks more comfortable.
Nb-hd 2.0 can save a lot of bandwidth costs, while also improving video viewing fluency, saving nearly twice the bit rate at high bit rate, saving 20%-30% at low bit rate, and achieving a total cost savings of over 30%. In addition, narrowband HD 2.0 can also improve sharpness, and if you want video sharpness, you can set the same bit rate.
In general, the better the video quality is, the higher the cost of bandwidth will be. As bandwidth is a high cost in video services, many developers are seeking to strike a balance between effect and cost. Ali Cloud’s narrowband HD technology can perfectly solve this problem, which is the embodiment of the practical value of cloud computing technology.
The original link
To read more articles, please scan the following QR code:
