实时音视频开发理论必备：如何省流量？视频高度压缩背后的预测技术

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文引用了“拍乐云Pano”的“深入浅出理解视频编解码技术”和“揭秘视频千倍压缩背后的技术原理之预测技术”文章部分内容，感谢原作者的分享。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"1、引言","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从 20 世纪 90 年代以来，数字音视频编解码技术迅速发展，一直是国内外研究的热点领域。随着5G的成熟和广泛商用，带宽已经越来越高，传输音视频变得更加容易。视频直播、视频聊天，已经完全融入了每个人的生活。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"视频为何如此普及呢？是因为通过视频能方便快捷地获取到大量信息。但视频数据量非常巨大，视频的网络传输也面临着巨大的挑战。于是视频编解码技术就出场了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"具体到实时视频场景，不仅仅是数据量的问题，实时通信对时延要求、设备适配、带宽适应的要求也非常高，要解决这些问题，始终离不开视频编解码技术的范畴。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文将从视频编解码技术的基础知识入手，引出视频编解码技术中非常基础且重要的预测技术，学习帧内预测和帧间预测的技术原理。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"2、相关文章","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"如果你是音视频技术初学者，以下3篇入门级干货非常推荐一读：","attrs":{}}]},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-2840-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"零基础，史上最通俗视频编码技术入门","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-3079-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"零基础入门：实时音视频技术基础知识全面盘点","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-3194-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"实时音视频面视必备：快速掌握11个视频技术相关的基础概念","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"3、为什么需要视频编解码","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，来复习一下视频编解码方面的理论常识。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"视频是由一系列图片按照时间顺序排列而成：","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1）每一张图片为一帧；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2）每一帧可以理解为一个二维矩阵；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"3）矩阵的每个元素为一个像素。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一个像素通常由三个颜色进行表达，例如用RGB颜色空间表示时，每一个像素由三个颜色分量组成。每一个颜色分量用1个字节来表达，其取值范围就是0~255。编码中常用的YUV格式与之类似，这里不作展开。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/7b/7b43136fad4e66920e5d09f3293d8592.png","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以1280x720@60fps的视频序列为例，十秒钟的视频有：1280*720*3*60*10 = 1.6GB。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如此大量的数据，无论是存储还是传输，都面临巨大的挑战。视频压缩或者编码的目的，也是为了保证视频质量的前提下，将视频减小，以利于传输和存储。同时，为了能正确还原视频，需要将其解码。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"PS：","attrs":{}},{"type":"text","text":"限于篇幅，视频编解码方面的技术原理就不在此展开，有兴趣强烈推荐从这篇深入学习：《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-2840-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十九）：零基础，史上最通俗视频编码技术入门","attrs":{}}]},{"type":"text","text":"》。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"总之，视频编解码技术的主要作用就是：在可用的计算资源内，追求尽可能高的视频重建质量和尽可能高的压缩比，以达到带宽和存储容量的要求。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为何突出“重建质量”？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因为视频编码是个有损的过程，用户只能从收到的视频流中解析出“重建”画面，它与原始的画面已经不同，例如观看低质量视频时经常会碰到的“块”效应。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如何在一定的带宽占用下，尽可能地保持视频的质量，或者在保持质量情况下，尽可能地减少带宽利用率，是视频编码的基本目标。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"用专业术语来说，即视频编解码标准的“率失真”性能：","attrs":{}}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"1）“率”是指码率或者带宽占用；","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2）“失真”是用来描述重建视频的质量。","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"与编码相对应的是解码或者解压缩过程，是将接收到的或者已经存储在介质上的压缩码流重建成视频信号，然后在各种设备上进行显示。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"4、什么是视频编解码标准","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"视频编解码标准，通常只定义上述的解码过程。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"例如 H.264 / AVC 标准，它定义了什么是符合标准的视频流，对每一个比特的顺序和意义都进行了严格地定义，对如何使用每个比特或者几个比特表达的信息也有精确的定义。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"正是这样的严格和精确，保证了不同厂商的视频相关服务，可以很方便地兼容在一起，例如用 iPhone、Android Phone 或者 windows PC 都可以观看同一在线视频网站的同一视频。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"世界上有多个组织进行视频编码标准的制定工作，国际标准组织 ISO 的 MPEG 小组、国际电信联盟 ITU-T 的 VCEG 小组、中国的 AVS 工作组、Google 及各大厂商组成的开放媒体联盟等。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"视频编码标准及发展历史：","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/64/641e270fd7bf780f8852f409f0924915.png","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"自 VCEG 制定 H.120标准开始，视频编码技术不断发展，先后成功地制定了一系列满足不同应用场景的视频编码标准。VCEG 组织先后制定了H.120、H.261、H.262(MPEG-2 Part 2)、H.263、H.263+、H.263++。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"MPEG也先后制定了MPEG-1、MPEG-2、MPEG-4 Part 2。以及两个国际组织合作制定的H.264/AVC、H.265/HEVC、H.266/VVC。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"中国自主知识产权的 AVS、AVS2、AVS3 视频编码标准；Google 制定的 VP8、VP9。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Google、思科、微软、苹果等公司组成的开放媒体联盟（AOM）制定的 AV1。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"这里特别提一下H.264/AVC：","attrs":{}},{"type":"text","text":"H.264/AVC虽有近20年历史，但它优秀的压缩性能、适当的运算复杂度、优秀的开源社区支持、友好的专利政策、强大的生态圈等多个方面的因素，依旧让它保持着强大的生命力，特别是在实时通信领域。像 ZOOM、思科 Webex 等视频会议产品和基于 WebRTC SDK 的视频服务，大多数主流场景都采用 H.264/AVC。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"有关视频编解码标准，这里就不深入展开。更多详细资料，可以读一下下面这些精选文章：","attrs":{}}]},{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-237-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（五）：认识主流视频编码技术H.264","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-266-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十三）：实时视频编码H.264的特点与优势","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-274-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十七）：视频编码H.264、VP8的前世今生","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-3028-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"爱奇艺技术分享：轻松诙谐，讲解视频编解码技术的过去、现在和将来","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}],"attrs":{}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"5、混和编码框架","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"纵观视频编解码标准历史，每一代视频标准都在率失真性能上有着显著的提升，他们都有一个核心的框架，就是基于块的混合编码框架（如下图所示）。它是由J. R. Jain 和A. K. Jain在1979年的国际图像编码学会(PCS 1979)上提出了基于块运动补偿和变换编码的混合编码框架。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/e4/e4e89e76883d806e924bb9479912f105.jpeg","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们一起来对该框架进行拆解和分析。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"从摄像头采集到的一帧视频：","attrs":{}},{"type":"text","text":"通常是 YUV 格式的原始数据，我们将它划分成多个方形的像素块依次进行处理（例如 H.264/AVC 中以16x16像素为基本单元），进行帧内/帧间预测、正变换、量化、反量化、反变换、环路滤波、熵编码，最后得到视频码流。从视频第一帧的第一个块开始进行空间预测，因当前正在进行编码处理的图像块和其周围的图像块有相似性，我们可以用周围的像素来预测当前的像素。我们将原始像素减去预测像素得到预测残差，再将预测残差进行变换、量化，得到变换系数，然后将其进行熵编码后得到视频码流。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"接下来：","attrs":{}},{"type":"text","text":"为了可以使后续的图像块可以使用已经编码过的块进行预测，我们还要对变换系统进行反量化、反变换，得到重建残差，再与预测值进行求合，得到重建图像。最后我们对重建图像进行环路滤波、去除块效应等，这样得到的重建图像，就可以用来对后续图像块进行预测了。按照以上步骤，我们依次对后续图像块进行处理。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"对于视频而言：","attrs":{}},{"type":"text","text":"视频帧与帧的间隔大约只有十到几十毫秒，通常拍摄的内容不会发生剧烈变化，它们之间存在非常强的相关性。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如下图所示，将视频图像分割成块，在时间相邻的图像之间进行匹配，然后将匹配之后的残差部分进行编码，这样可以较好地去除视频信号中的视频帧与帧之间的冗余，达到视频压缩的目的。这就是运动补偿技术，直到今天它仍然是视频编解码的核心技术之一。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"运动估计和运动补偿：","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/b0/b0fa5aaf1e61dafb9e08208cfba182b5.jpeg","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"变换编码的核心思想：","attrs":{}},{"type":"text","text":"是把视频数据分割成块，利用正交变换将数据的能量集中到较少几个变换系数上。结合量化和熵编码，我们可以获得更有效的压缩。视频编码中信息的损失和压缩比的获得，很大程度上来源于量化模块，就是将源信号中的单一样本映射到某一固定值，形成多到少的映射，从而达到压缩的目的，当然在压缩的过程中就引入了损失。量化后的信号再进行无损的熵编码，消除信号中的统计冗余。熵编码的研究最早可以追溯到 20 世纪 50 年代，经过几十年的发展，熵编码在视频编码中的应用更加成熟、更加精巧，充分利用视频数据中的上下文信息，将概率模型估计得更加准确，从而提高了熵编码的效率。例如H.264/AVC中的Cavlc（基于上下文的变长编码）、Cabac（基于上下文的二进制算术编码）。算术编码技术在后续的视频编码标准，如AV1、HEVC/H.265、VVC/H.266 中也有应用。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"视频编码发展至今，VVC/H.266 作为最新制定的标准，采纳了一系列先进的技术，对混合编码框架的各个部分都进行了优化和改进，使得其率失真性能相比前一代标准，又提高了一倍。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"例如：","attrs":{}},{"type":"text","text":"VVC/H.266 采用了128x128大小的基本编码单元，并且可以继续进行四叉树划分，支持对一个划分进行二分、三分；色度分量独立于亮度分量，支持单独进行划分；更多更精细的帧内预测方向、帧间预测模式；支持多种尺寸和形式的变换、环内滤波等。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"VVC/H.266 的制定，目标是对多种视频内容有更好支持，例如屏幕共享内容、游戏、动漫、虚拟现实内容（VR、AR）等。其中也有特定的技术被采纳进标准，例如调色板模式、帧内运动补偿、仿射变换、跳过变换、自适应颜色变换等。   ","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"回到本文的正题，接下来的内容，我们着重介绍视频编解码中的预测技术。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"6、帧内预测技术","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"视频数据被划分成方块之后，相邻的方块的像素，以及方块内的像素，颜色往往是逐渐变化的，他们之间有比较强的有相似性。这种相似性，就是空间冗余。既然存在冗余，就可以用更少的数据量来表达这样的特征。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"比如：","attrs":{}},{"type":"text","text":"先传输第一个像素的值，再传输第二个像素相对于第一个像素的变化值，这个变化值往往取值范围变小了许多，原来要8个bit来表达的像素值，可能只需要少于8个bit就足够了。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同样的道理，以像素块为基本单位，也可以进行类似的“差分”操作。我们从示例图中，来更加直观地感受一下这样的相似性。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/12/12738fd8e3671706f133fb07b338db94.jpeg","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"如上图中所标出的两个8x8的块：","attrs":{}},{"type":"text","text":"其亮度分量（Y）沿着“左上到右下”的方向，具有连续性，变化不大。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"假如：","attrs":{}},{"type":"text","text":"我们设计某种特定的“模式”，使其利用左边的块来“预测”右边的块，那么“原始像素”减去“预测像素”就可以减少传输所需要的数据量，同时将该“模式”写入最终的码流，解码器便可以利用左侧的块来“重建”右侧的块。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"极端一点讲：","attrs":{}},{"type":"text","text":"假如左侧的块的像素值经过一定的运算可以完全和右侧的块相同，那么编码器只要用一个“模式”的代价，传输右侧的块。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，视频中的纹理多种多样，单一的模式很难对所有的纹理都适用，因此标准中也设计了多种多样的帧内预测模式，以充分利用像素间的相关性，达到压缩的目的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"例如下图所示的H.264中9种帧内预测方向：","attrs":{}},{"type":"text","text":"以模式0（竖直预测）为例，上方块的每个像素值（重建）各复制一列，得到帧内预测值。其它各种模式也采用类似的方法，不过，生成预测值的方式稍有不同。有这么多的模式，就产生了一个问题，对于一个块而言，我们应该采用哪种模式来进行编码呢？最佳的选择方式，就是遍历所有的模式进行尝试，计算其编码的所需的比特数和产生的质量损失，即率失真优化，这样明显非常复杂，因而也有很多种其它的方式来推断哪种模式更好，例如基于SATD或者边缘检测等。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从H.264的9种预测模式，到AV1的56种帧内方向预测模式，越来越多的模式也是为了更加精准地预测未编码的块，但是模式的增加，一方面增加了传输模式的码率开销，另一方面，从如此重多的模式中选一个最优的模式来编码，使其能达到更高的压缩比，这对编码器的设计和实现也提出了更高的要求。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/4e/4e6e2d0f25ddc5b67ca3f7e28e4708c0.png","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"7、帧间预测技术","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"以下5张图片是一段视频的前5帧：","attrs":{}},{"type":"text","text":"可以看出，图片中只有Mario和砖块在运动，其余的场景大多是相似的，这种相似性就称之为时间冗余。编码的时候，我们先将第一帧图片通过前文所述的帧内预测方式进行编码传输，再将后续帧的Mario、砖块的运动方向进行传输，解码的时候，就可以将运动信息和第一帧一起来合成后续的帧，这样就大大减少了传输所需的bit数。这种利用时间冗余来进行压缩的技术，就是运动补偿技术。该技术早在H.261标准中，就已经被采用。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/6c/6c0fb536ab3cb13509e0b5b29f89bba9.png","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"细心地读者可能已经发现：","attrs":{}},{"type":"text","text":"Mario和砖块这样的物体怎么描述，才能让它仅凭运动信息就能完整地呈现出来？","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其实视频编码中并不需要知道运动的物体的形状，而是将整帧图像划分成像素块，每个像素块使用一个运动信息。即基于块的运动补偿。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下图中红色圈出的白色箭头即编码砖块和Mario时的运动信息，它们都指向了前一帧中所在的位置。Mario和砖块都有两个箭头，说明它们都被划分在了两个块中，每一个块都有单独的运动信息。这些运动信息就是运动矢量。运动矢量有水平和竖直两个分量，代表是的一个块相对于其参考帧的位置变化。参考帧就是已经编码过的某一（多）个帧。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/cd/cdc8cec5cb67076c1edfee50e128f45a.png","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"当然：","attrs":{}},{"type":"text","text":"传输运动矢量本身就要占用很多 bit。为了提高运动矢量的传输效率，主要有以下措施。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"一方面：","attrs":{}},{"type":"text","text":"可以尽可能得将块划分变大，共用一个运动矢量，因为平坦区域或者较大的物体，他们的运动可能是比较一致的。从 H.264 开始，可变块大小的运动补偿技术被广泛采用。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"另一方面：","attrs":{}},{"type":"text","text":"相邻的块之间的运动往往也有比较高的相似性，其运动矢量也有较高的相似性，运动矢量本身也可以根据相邻的块运动矢量来进行预测，即运动矢量预测技术；","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"最后：","attrs":{}},{"type":"text","text":"运动矢量在表达物体运动的时候，有精度的取舍。像素是离散化的表达，现实中物体的运动显然不是以像素为单位进行运动的，为了精确地表达物体的运动，需要选择合适的精度来定义运动矢量。各视频编解码标准都定义了运动矢量的精度，运动矢量精度越高，越能精确地表达运动，但是代价就是传输运动矢量需要花费更多的bit。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"H.261中运动矢量是以整像素为精度的，H.264中运动矢量是以四分之一像素为精度的，AV1中还增加了八分之一精度。一般情况，时间上越近的帧，它们之间的相似性越高，也有例外，例如往复运动的场景等，可能相隔几帧，甚至更远的帧，会有更高的相似度。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了充分利用已经编码过的帧来提高运动补偿的准确度，从H.264开始引入了多参考帧技术。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"即：","attrs":{}},{"type":"text","text":"一个块可以从已经编码过的很多个参考帧中进行运动匹配，将匹配的帧索引和运动矢量信息都进行传输。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那么如何得到一个块的运动信息呢？最朴素的想法就是，将一个块，在其参考帧中，逐个位置进行匹配检查，匹配度最高的，就是最终的运动矢量。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"匹配度：常用的有SAD（Sum of Absolute Difference）、SSD（Sum of Squared Difference）等。逐个位置进行匹配度检查，即常说的全搜索运动估计，其计算复杂度可想而知是非常高的。为了加快运动估计，我们可以减少搜索的位置数，类似的有很多算法，常用的如钻石搜索、六边形搜索、非对称十字型多层次六边形格点搜索算法等。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"以钻石搜索为例，如下图所示，以起始的蓝色点为中心的9个匹配位置，分别计算这9个位置的SAD，如果SAD最小的是中心位置，下一步搜索中心点更近的周围4个绿色点的SAD，选择其中SAD最小的位置，继续缩小范围进行搜索；如果第一步中SAD最小的点不在中心，那么以该位置为中心，增加褐色的5或者3个点，继续计算SAD，如此迭代，直到找到最佳匹配位置。","attrs":{}}]},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/95/9545a2196893e253b20b16087bce7c74.png","alt":"","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"编码器在实现时，可根据实际的应用场景，对搜索算法进行选择。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"例如：","attrs":{}},{"type":"text","text":"在实时音视频场景下，计算复杂度是相对有限的，运动估计模块要选择计算量较小的算法，以平衡复杂度和编码效率。当然，运动估计与运动补偿的复杂度还与块的大小，参考帧的个数，亚像素的计算等有关，在此不再深入展开。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"更多预测技术方面的原理这里就不再赘述。如果你对上面所述的预测技术理解上感到力不从心，这里有篇入门级的文章，可以先读读这篇《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-235-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（四）：视频编解码之预测技术介绍","attrs":{}}]},{"type":"text","text":"》。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"8、写在最后","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"音视频编解码技术，归根结底就是在有限的资源下（网络带宽、计算资源等），让音质更清晰、视频更高质。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这其中，对于视频来说，质量的提升仍然有很多可以深入研究的热点问题。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"比如：","attrs":{}},{"type":"text","text":"基于人眼的主观质量优化，主要利用人眼的视觉特性，将掩蔽效应、对比度灵敏度、注意力模型等与编码相结合，合理分配码率、减少编码损失引起的视觉不适。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"AI在视频编解码领域的应用：包括将多种人工智能算法，如分类器、支持向量机、CNN等对编码参数进行快速选择，也可以使用深度学习对视频进行编码环外与编码环内的处理，如视频超分辨率、去噪、去雾、自适应动态范围调整等编码环外处理，达到提升视频质量的目的。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外还有打破传统混合编码框架的深度神经网络编码，如Nvidia的Maxine视频会议服务，利用深度学习来提取特征，然后对特征进行传输以节省带宽。","attrs":{}}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"附录：参考资料","attrs":{}}]},{"type":"blockquote","content":[{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-228-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（一）：视频编解码之理论概述","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-229-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（二）：视频编解码之数字视频介绍","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-232-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（三）：视频编解码之编码基础","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-235-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（四）：视频编解码之预测技术介绍","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-237-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（五）：认识主流视频编码技术H.264","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-241-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（六）：如何开始音频编解码技术的学习","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-242-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（七）：音频基础及编码原理入门","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-243-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（八）：常见的实时语音通讯编码标准","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-247-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（九）：实时语音通讯的回音及回音消除概述","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-250-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十）：实时语音通讯的回音消除技术详解","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-251-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十一）：实时语音通讯丢包补偿技术详解","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-253-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十二）：多人实时音视频聊天架构探讨","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-266-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十三）：实时视频编码H.264的特点与优势","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-267-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十四）：实时音视频数据传输协议介绍","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-269-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十五）：聊聊P2P与实时音视频的应用情况","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-270-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十六）：移动端实时音视频开发的几个建议","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-274-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十七）：视频编码H.264、VP8的前世今生","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-2230-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十八）：详解音频编解码的原理、演进和应用选型","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-2840-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"即时通讯音视频开发（十九）：零基础，史上最通俗视频编码技术入门","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-853-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"移动端实时音视频直播技术详解（一）：开篇","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-955-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"移动端实时音视频直播技术详解（二）：采集","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-960-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"移动端实时音视频直播技术详解（三）：处理","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-965-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"移动端实时音视频直播技术详解（四）：编码和封装","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-967-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"移动端实时音视频直播技术详解（五）：推流和传输","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"《","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-972-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"移动端实时音视频直播技术详解（六）：延迟优化","attrs":{}}]},{"type":"text","text":"》","attrs":{}}]}]}],"attrs":{}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong","attrs":{}}],"text":"本文已同步发布于“即时通讯技术圈”公众号。","attrs":{}}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"同步发布链接是：","attrs":{}},{"type":"link","attrs":{"href":"http://www.52im.net/thread-3581-1-1.html","title":null,"type":null},"content":[{"type":"text","text":"http://www.52im.net/thread-3581-1-1.html","attrs":{}}]}]}],"attrs":{}}]}