Shulou(Shulou.com)06/03 Report--

Author: Jiang Sheng. He is now the Senior Technical Manager of PPTV& SUNING Sports Codec Kernel Development. Background in image coding and decoding, high performance computing; engaged in H.264 decoder development and optimization, H.265 encoder development and optimization. More than 10 years experience in audio and video media development. Currently responsible for the design and R & D of H.265 encoder architecture, undertake video business within SUNING Group, promote the architecture improvement of video cloud platform and enhance service capability.

Short video is a new expression of information, which combines text, voice and video. It can more intuitively and stereoscopically meet the expression of users, communication needs, and meet the needs of display and sharing between people.

In recent years, short videos have shown a blowout development. In live broadcast websites, copyright and bandwidth constitute the vast majority of the cost, and the cost of bandwidth is second only to copyright. For user-generated short video platforms, the proportion of bandwidth costs may be higher. It is conceivable that bandwidth costs may rise exponentially under the stimulation of 4K or even 6K video and 5G. The future video platform may not be able to bear the dilemma of high bandwidth costs. So saving bandwidth has become an important demand point. This in-depth disclosure of PPTV& SUNING Sports to improve H.265 video compression ratio of the technical selection and tuning methods.

The present situation and characteristics of short Video

Short video introduction

Short video is a kind of expression of information, which is no more than 15 minutes in length. It mainly relies on mobile intelligent terminal to achieve fast shooting and beautification editing, and can be shared in real time and seamlessly docked on social media platform. It combines text, voice and video, and can more intuitively and stereoscopically meet the expression of users, communication needs, and meet the demands of display and sharing among people.

Operation and maintenance cost of short video

In live broadcast websites, copyright and bandwidth constitute the vast majority of the cost, and the cost of bandwidth is second only to copyright. For user-generated short video platforms, the proportion of bandwidth costs may be higher. It is conceivable that bandwidth costs may rise exponentially under the stimulation of 4K or even 6K video and 5G. The future video platform may not be able to bear the dilemma of high bandwidth costs.

Ways to reduce costs

P2P CDN: since the cost of bandwidth remains high, let users share bandwidth through P2P to achieve the purpose of reducing bandwidth cost. P2P CDN is based on this principle. The technology behind many "making money" boxes on the market is P2PCDN. This is why there are so many blockchain projects on the market aimed at bandwidth sharing, all of which are backed by P2P CDN.

Encoder: at present, H.264 Codec is the most popular HEVC in the network. Compared with H.264 HEVC, it has more than 30% bandwidth savings. However, trapped by complex patent licensing, HEVC is facing a challenge from the AOM alliance-AV1. Facebook tests show that AV1 has a compression ratio of at least 30% higher than x264 and can compete with HEVC. However, due to its high complexity, AV1 still has a long way to go from practical application.

Performance advantages of H.265 and H.264

Bit rate savings and clarity

Introduction of H265 coding architecture and tools

Structural differences between H.265 and H.264

The difference of coding framework between H265 and H264

H.265 adopts hybrid coding and decoding, and its structure is basically the same as that of H264. The details are as follows:

Coding block partition structure: using CU, PU, TU recursive structure

Parallel tools: add parallel tools such as Tile and WPP to improve speed

Filter: added SAO module after deblocking filter

Coding block partition structure: CU, PU, TU recursive structure:

Contribution of SAO to compression ratio

PPTV& SUNING Sports rate reduction Technology

Algorithm 1: sequence characteristic analysis

Sequence characteristic Analysis & Architecture Diagram

Algorithm 1: analysis of time period characteristics

In the actual coding process, the transcoding server usually sets a fixed bit rate for each type of video, but in fact, in the same kind of video, the details and motion characteristics of each clip sequence are different, and the required bit rate is different under the same definition. In order to overcome this problem, PPTV& SUNING Sports dynamically allocates the bit rate according to the characteristics of each video. To achieve the purpose of saving code rate.

Implementation method:

The complexity of the video is estimated according to the average QP obtained by pass1 and the QP obtained by this level of complex video coding, and then the final target bit rate of pass2 is determined according to the video complexity and the current set target bit rate. This algorithm can effectively reduce the video bit rate without changing the flow of the rate allocation algorithm.

Feasibility analysis:

Analysis: the QP obtained by pass1 with different content, same resolution and different source chip bit rate varies with the target bit, as shown on the left. As can be seen from the figure, the QP obtained by pass1 varies greatly with the target bit under the same source chip bit rate of different video content, and the distribution of the sequence is basically the same under different source chip bit rate, and the more complex video sequence with the same source chip bit rate, the larger the corresponding QP after pass1 coding, which verifies the correctness of our algorithm hypothesis.

Bitrate-QP curve fitting:

Video complexity diagram

Test the most complex 3, 5, and 8 video sequences. Then the best model is obtained by fitting it with matlab.

1920x1080-3 indicates that the resolution of video sequence coding is 1920x1080, and the model is fitted by the three most complex video sequences of 25 video sequences, and the others are similar.

Curve fitting effect

In the case of high resolution, the complexity curve of 3, 5 and 8 most complex video sequences is very close, while in the case of low resolution, the complexity curve of 3, 5 and 8 most complex video sequences is very close to that below 3000 kbps.

Set bit rate

Now, what needs to be solved is how to allocate the appropriate target bit rate according to the most complex video sequence complexity when the target bit rate is set to k, the quantization parameters obtained from video sequence pass1 coding. A total of 10 models are proposed, of which the following models are the best:

Bit rate reduction effect

Prospects for the future

PPTV& SUNING Sports Division will introduce the subjective score of the image into the encoder in the future as a reference factor for rate control, so that the image quality and code rate are well controlled.

SUNING Video Cloud, a subsidiary of Suning, has served more than 2000 customers. With ten years of PPTV media technology and service experience, SUNING Video Cloud is an one-stop SaaS service platform focused on video field built by combining streaming media technology, P2P, CDN distribution, mass storage, security strategy, etc. SUNING Video integrates video LVB, cloud VOD, cloud upload, cloud transcoding, cloud storage, cloud statistics and other functions, and supports customers' business needs of various video scenarios on multiple platforms.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.