基于概率模型的3D-HEVC中依赖视图早期合并模式决策优化

14 浏览量更新于2024-07-14 收藏 1.07MB PDF 举报

本文主要探讨了在3D-HEVC（High Efficiency Video Coding for Three-Dimensional Videos）标准的扩展中，针对依赖视图的编码效率优化问题。3D-HEVC旨在提升多视点视频的压缩效率，它继承了HEVC中的预测模式，但在处理关联视图时，需要同时进行运动估计（ME）和深度估计（DE），这无疑增加了计算负担。为了降低这种复杂性并提高编码效率，作者提出了一种基于概率模型的早期合并模式决策方法。首先，该方法利用先前编码块的层次结构和纹理相关性，构建了先验概率模型。通过分析已编码区域的统计特性，研究人员设计了一个模型，能够预测当前块的纹理模式和可能的预测方式，从而减少不必要的ME和DE操作。这种方法旨在利用历史数据的统计规律来指导早期决策，提前确定哪些视图或深度图之间的模式可以合并，从而节省编码过程中的计算资源。其次，文章构建了后验概率模型，这是通过对当前编码块的Coded Block Flag (CBF)进行分析实现的。CBF是一个指示当前块是否被编码的标志，通过对CBF的观察，可以推测出该块与其他块的相似度，进一步用于判断是否适合进行早期合并。后验概率模型是在先验模型基础上结合编码结果的实时反馈，提高了决策的准确性。最后，结合先验和后验概率模型，研究人员提出了一个联合模型，将两者的优势结合起来。这个联合模型不仅考虑了历史数据的规律，也考虑了当前编码状态的信息，从而实现了更为精确的早期合并模式决策。这种方法有望在保持编码效率的同时，显著降低3D-HEVC中依赖视图编码的计算复杂度，对于实际应用具有重要的理论和实践价值。这篇文章深入研究了基于概率模型的3D-HEVC中从属视图的早期合并模式决策策略，通过有效地利用统计信息和编码反馈，为解决多视点视频编码中的性能与计算负载之间的平衡问题提供了一种创新解决方案。这对于推动3D视频编码技术的发展，尤其是在资源受限的设备上实现高效视频编码具有重要意义。

Probability Model-Based Early Merge Mode Decision for Dependent Views Coding 85:3

because 3D-HEVC supports more exible quad-tree coding structures and prediction techniques

than previous 3D video coding standards [6, 8].

In recent years, a few fast mode decision methods have been proposed at each CU depth under

the framework of either HEVC or 3D-HEVC. These include correlation-based methods which ex-

ploit mode correlation, spatial-temporal correlation, interview correlation, RD cost correlation,

hierarchical correlation, Motion Vector (MV) and Coded Block Flag (CBF), among others. The

correlation-based methods are always built based on observations and some experimental sta-

tistics. They have the advantages of implementation simplicity and require less modications to

the encoder. For example, in Shen et al. [20], a fast intermode decision approach was proposed for

HEVC by jointly using interlevel correlation, spatiotemporal correlation, MV, and RD cost corre-

lation. In Jung and Park [5], a fast mode decision method was proposed using the RD cost and bit

cost. In Zhao et al. [42], a hierarchical structure-based fast mode decision algorithm was proposed

by using the colocated depth information from a previous frame to predict the split structure of

the current block. In Zhang et al. [34], an ecient fast mode decision method was proposed for the

interprediction of HEVC by exploiting the relationship between impossible modes and the distri-

bution of distortions to avoid checking unnecessary modes. In Hu et al. [4], a fast mode decision

algorithm was proposed based on the Neyman-Pearson rule to balance RD performance loss and

complexity reduction, which consists of an early SKIP mode decision and a fast CU size decision.

Since the independent views of 3D-HEVC are independently encoded by the HEVC-based codec,

these fast methods of HEVC are suitable for them. In Zhang et al. [36], an ecient multiview video

plus depth coding scheme was proposed for 3D-HEVC based on the complexity classication of

a treeblock. In Zhang et al. [38], a fast depth map mode decision algorithm was proposed for 3D-

HEVC by jointly using the correlation of a depth map-texture video and the edge information of

a depth map. In Shen et al. [16], a fast mode decision algorithm was proposed for 3D-HEVC by

jointly exploiting the interview coding mode correlation, the intercomponent correlation, and the

interlevel correlation in the quadtree structure.

However, these approaches do not fully exploit the early Merge mode decision, which does not

require time-consuming ME and DE before checking other modes. That is, if the Merge mode can

be terminated early, the encoding complexity will be signicantly reduced by skipping the remain-

ing modes that have complex ME and DE. In Yang et al. [31], an early SKIP mode decision method

was proposed by rst checking the Inter_2N×2N and the Merge modes. All the other modes in the

current CU depth are skipped if the prediction results from the current CU satisfy the condition

that both Motion Vector Dierence (MVD) and the residuals of Inter_2N×2N mode are zero. The

SKIP mode is the special case of the Merge mode in which neither performs ME nor encodes the

residuals. In Li et al. [9], a unimodal stopping model was established for an early SKIP mode de-

cision by exploiting RD cost and hierarchical mode correlations. In Pan et al. [13], an early Merge

mode decision method was proposed based on the All-Zero Block (AZB), hierarchical correlation,

and the ME information of the Inter_2N×2N mode. In Tariq et al. [23], an early Merge mode deci-

sion algorithm was proposed for HEVC based on spatial/temporal motion consistency. Likewise,

some early Merge mode methods were proposed for 3D-HEVC. In Zhang et al. [37], an early SKIP

mode decision algorithm was proposed for 3D-HEVC by exploiting spatial and interview corre-

lations. In Zhang et al. [35], an early Merge mode decision method was proposed for dependent

texture views by exploiting interview correlation, which is now adopted by 3D-HEVC. In Song

and Jia [29], an early Merge mode decision scheme was proposed for dependent texture coding

in 3D-HEVC by exploiting the interview correlation and the hierarchical correlation among depth

levels “2” and “3.” Similarly, the interview correlation was exploited for dependent depth maps

coding in Chen et al. [2]. Though these early Merge mode decision methods signicantly reduce

ACM Trans. Multimedia Comput. Commun. Appl., Vol. 14, No. 4, Article 85. Publication date: September 2018.

剩余14页未读，继续阅读

weixin_38723192

粉丝: 8
资源: 870

基于概率模型的3D-HEVC中依赖视图早期合并模式决策优化

Test Model 11 of 3D-HEVC and MV-HEVC： JVT3V-K1003

MV-HEVC and 3D-HEVC Reference Software 16.2

论文研究-基于3D-HEVC深度建模模式的快速模式判决的研究 .pdf

MV-HEVC中基于视图依赖的帧级比特分配优化方法

基于学习模型的3D-HEVC提前Merge模式优化算法：编码效率与时间降低41.9%

优化3D-HEVC编码的快速模式决策算法

Test Model 11 of 3D-HEVC and MV-HEVC.docx

基于贝叶斯决策规则的3D-HEVC快速在线学习参数决策算法

3D-HEVC中英文对照

3D-HEVC中深度图帧内预测模式判决过程的改进

最新资源