HEVC编码标准HM10参考程序详解

需积分: 10 184 浏览量更新于2024-07-24 收藏 2.07MB DOC 举报

"HEVC参考程序10.0是JCTVC（ITU-T SG16 WP3 和 ISO/IEC JTC1 SC29 WG11 联合视频编码团队）为了研发新一代视频编码标准HEVC（也称为H.265）而创建的一个编码端实现。这个文档是JCTVC-L1002_v23，由Il-Koo Kim（三星）和Ken McCann（ZetaCast）等人共同撰写，作为JCT-VC的输出文档，用于报告HEVC测试模型10（HM10）编码器的详细信息。" HEVC（高效视频编码，High Efficiency Video Coding）是视频编码技术领域的一项重要进展，旨在提供比前一代标准H.264/AVC更高的压缩效率，从而在有限的带宽下传输更高质量的视频内容。HEVC参考程序10.0（HM10）是这一标准的一个关键组成部分，它提供了编码器的设计和实现，帮助研究人员和开发者理解HEVC的编码流程和技术细节。在HM10中，编码器的描述涵盖了以下几个核心方面： 1. **编码结构概述**：HEVC的编码结构通常包括宏块、编码单元（CU）、预测单元（PU）和变换单元（TU）。宏块的概念在HEVC中被逐步细化，通过CU来适应不同大小和形状的编码区域，以提高编码效率。这种分层结构允许更加灵活地处理图像内容，优化压缩性能。 2. **图片分区**：在HEVC中，图片被划分为不同的编码单元，这些单元可以独立编码并具有可变大小。这种分区方式使得编码器可以根据图像内容的复杂性进行自适应编码，进一步提升压缩效果。此外，HEVC引入了四叉树分割和二叉树分割，使得编码单元的划分更加精细，有助于提高编码质量。 3. **预测技术**：HEVC采用多种预测模式，包括帧内预测和帧间预测。帧内预测利用同一帧内的相邻像素来预测当前像素值；帧间预测则依赖于之前或之后的帧，通过运动补偿来预测像素值。HM10编码器详细描述了这些预测技术的实现。 4. **变换与量化**：HEVC使用离散余弦变换（DCT）和量化过程来减少信号的统计冗余。量化参数可以调整，以平衡编码质量和码率。HM10编码器会详细阐述这一过程。 5. **熵编码**：熵编码如上下文自适应二进制算术编码（CABAC）或上下文自适应变量长度编码（CAVLC）用于将编码后的数据转换成更紧凑的二进制流，便于传输和存储。 6. **循环滤波器**：HEVC引入了增强的去方块滤波器和多向运动补偿滤波器，以改善编码后的图像质量，特别是在低比特率下。 7. **工具集**：HEVC引入了一系列高级编码工具，如多参考帧、分块运动估计、深度信息处理、自适应色彩空间转换等，以提高编码效率。通过HEVC参考程序10.0，开发者和研究人员可以深入理解HEVC编码流程的各个方面，这对于标准的改进、优化以及在实际应用中的实现至关重要。这份文档不仅提供了理论性的介绍，还包含了编码器的具体实现细节，对于那些希望在HEVC基础上进行二次开发或研究的人来说，是一份非常宝贵的资源。

(HM1) [2]. The majority of the tools within HM1 were in the TMuC, but HM1 had substantially fewer coding tools and
hence there  was  a substantial decrease in the computational resources necessary for encoding and decoding.  Further
optimizations of the HEVC Test Model in HM2 [3], HM3 [4], HM4 [5], HM5 [6], HM6 [7], HM7[8], HM8[9], HM9[10]
and HM10 were specified at subsequent meetings, with each successive model achieving better performance than the
previous in terms of the trade-off between coding efficiency and complexity.
2 Scope
This document provides an encoder-side description of the HEVC Test Model (HM), which serves as a tutorial on the
encoding model implemented in the  HM software. The purpose of this text is to share a common understanding on
reference encoding methods supported in the HM software, in order to facilitate the assessment of the technical impact of
proposed new technologies during the HEVC standardization process. Although brief descriptions of the HEVC design
are provided to help understanding of the HM, the corresponding sections of the HEVC draft specification [11] should be
referred  to for   any  descriptions regarding   normative   processes.  A further   document  [13]  defines the   common   test
conditions and software reference configurations that should be used for experimental work.
3 Description of HEVC Test Model
3.1 General overview of coding structure
The   HEVC   standard   is   based   on   the   well-known   block-based   hybrid   coding   architecture,   combining   motion-
compensated prediction and transform coding with high-efficiency  entropy coding. However, in contrast to  previous
video coding standards, it employs a flexible quad-tree coding block partitioning structure that enables the use of large
and multiple sizes of  coding, prediction, and transform blocks. It also employs  improved intra prediction and coding,
adaptive motion parameter prediction and coding, a new loop filter and an enhanced version of context-adaptive binary
arithmetic coding (CABAC) entropy coding. New high level structures for parallel processing are also employed. 
A general block-diagram of the HM encoder is depicted in Figure   3 -1 below.
Figure 3-1 – Simplified block diagram of HM encoder

The picture partitioning structure is described in Section 3.2. The input video is first divided into blocks called coding

tree units (CTUs), which perform a role that is broadly analogous to that of macroblocks in previous standards. The

coding unit (CU) defines a region sharing the same prediction mode (intra, inter or skip) and it is represented by the leaf

node of a quadtree structure. The prediction unit (PU) defines a region sharing the same prediction information. The

transform unit (TU), specified by another quadtree, defines a region sharing the same transformation and quantization.

The intra-picture coding prediction processes are described in Section 3.3. The best intra mode among a total of 35

modes (Planar, DC and 33 angular directions) is selected and coded. Mode dependent context sample smoothing is

applied to increase prediction efficiency and the three most probable modes (MPM) are used to increase symbol coding

efficiency.

The inter-picture prediction processes are described in Section 3.4. The best motion parameters are selected and coded by

merge mode and adaptive motion vector prediction (AMVP) mode, in which motion predictors are selected and

explicitly coded among several candidates. To increase the efficiency of motion-compensated prediction, a non-cascaded

interpolation structure with 1D FIR filters is used. An 8-tap or 7-tap filter is directly applied to generate the values of

half-pel and quarter-pel luma samples, respectively. A 4-tap filter is utilized for chroma interpolation.

Transforms and quantization are described in Section 3.5. Residuals generated by subtracting the prediction from the

input are spatially transformed and quantized. In the transform process, matrices which are approximations to DCT are

used. A partial butterfly structure is implemented for the transformation, to reduce computational cost. In the case of 4x4

intra predicted residuals, an approximation to DST is used for the luma. 52-level quantization steps and rate-distortion

optimized quantization (RDOQ) are used in the quantization process. Reconstructed samples are created by inverse

quantization and inverse transform.

Entropy coding is described in Section 3.6. It is applied to the generated symbols and quantized transform coefficients in

the encoding process using a Context-based Adaptive Binary Arithmetic Coding (CABAC) process.

Loop filtering is described in Section 3.7. After reconstruction, two in-loop filtering processes are applied to achieve

better coding efficiency and visual quality: deblocking filtering and sample adaptive offset (SAO). Reconstructed CTUs

are assembled to construct a picture that is stored in the decoded picture buffer to be used to encode the next picture of

input video.

3.2 Picture partitioning

3.2.1 CTU partitioning

Pictures are divided into a sequence of coding tree units (CTUs). The CTU concept is broadly analogous to that of the

macroblock in previous standards such as AVC [14]. For a picture that has three sample arrays, a CTU consists of an

NxN block of luma samples together with two corresponding blocks of chroma samples.

The maximum allowed size of the luma block in a CTU is specified to be 64x64 in Main profile.

Figure 3-2 – Example of a picture divided into CTUs

剩余35页未读，继续阅读

zyjohnyoung

粉丝: 0
资源: 3

HEVC编码标准HM10参考程序详解

HEVC参考程序10.0版本

HEVC 参考代码（HM10.0）

HEVC 参考代码 HM10.0

HEVC参考测试模型 HM10.0

HEVC参考程序11.0版本

HEVC最新参考模型HM10.0

高效视频编码（HEVC）实现视频编码

HEVC帧内视频编码教程（HEVC视频编码技术的研究与实现）

H.265/HEVC 下一代视频编码标准 参考软件实现

HEVC参考软件

最新资源

H.265/HEVC 下一代视频编码标准参考软件实现