3D-HEVC预测编码算法研究

发布时间：2019-01-30 15:29

【摘要】：三维视频(3DV)可为观众带来身临其境的立体感觉,是近期视频编码领域的一个研究点。多视点视频加深度(Multiview video plus depth,MVD)因其具有后向兼容、支持交互等优点被运动图像专家组(MPEG)选作新一代3DV标准的数据格式。由于多视点视频信息和多视点深度信息均需要编码传输至解码端,使得码率明显增加,因此提高3DV编码效率显得尤为重要。三维高效率视频编码(Three dimension high efficiency video coding,3D-HEVC)是MEPG正在孵化的标准,其在HEVC的基础上研究高效编码MVD。3D-HEVC针对3DV的特点提出了多项编码工具,新增约50%的码率提供立体视觉体验。本文重点研究用于提高3DV编码效率的先进残差预测技术及利用视点间相关性的3DV快速编码技术。具体研究内容如下:先进残差预测技术(advanced residual prediction,ARP)是基于视点间的残差相关性提出的一种3D视频编码技术,该技术可有效降低非基本视点的编码码率。其关键点在于利用已有运动信息为当前残差构造残差预测值以进一步预测。若基本视点对应块采用帧内编码块,则其不包含运动信息,这一定程度上限制了先进残差预测的使用。针对上述问题,本文提出了一种改进的先进残差算法。该算法利用帧内编码块的相邻块运动矢量代替帧内编码块的零矢量,以提高预测精度,改善ARP的编码效率。实验结果表明,采用本文所提算法,非基本视点的编码码率平均降低了0.12%。对于多视点视频,同一时刻不同视点的视频在图像内容上具有非常强的相关性。3D-HEVC中采用了精细预测单元划分算法以提高编码效率。本文基于视点间相关性提出了一种帧间预测单元优化算法以降低3D-HEVC编码复杂度。根据视点间PU模式之间的相关性,对当前PU的划分模式构建候选列表,精简PU候选模式。在3D-HEVC的变换部分,TU采用残差四叉树变换结构,通过遍历比较不同深度的变换单元的率失真代价得到最佳TU划分。残差四叉树变换结构在提升编码效率的同时,计算复杂度显著增加。而3D-HEVC中提出的先进残差预测技术充分利用残差的相关性,降低了残差能量。实验统计发现约97.38%的使用先进残差预测技术的TU块并未进行划分。基于此,本文通过定义使用先进残差预测技术的标志位,提出了提前终止TU划分优化算法。实验结果表明,在编码效率和峰值信噪比损失很小的情况下,相比于3D-HEVC公共测试平台,本文所提两种优化算法对非基本视点的纹理图像的编码时间平均减少32.2%。
[Abstract]:3D video (3DV) is a research point in the field of video coding. Multi-view Video with depth (Multiview video plus depth,MVD) has been chosen as the new data format of 3DV standard by (MPEG) because of its backward compatibility and support for interaction. Since both the multi-view video information and the multi-view depth information need to be encoded and transmitted to the decoding end, the bit rate is obviously increased, so it is particularly important to improve the efficiency of 3DV coding. Three dimensional high efficiency video coding (Three dimension high efficiency video coding,3D-HEVC) is a standard that MEPG is hatching. Based on HEVC, the efficient coding MVD.3D-HEVC is studied and many coding tools are put forward according to the characteristics of 3DV. Add about 50% bit rate to provide stereoscopic vision experience. This paper focuses on the advanced residual prediction technology used to improve the efficiency of 3DV coding and the 3DV fast coding technology which utilizes the interview correlation. The main contents of this paper are as follows: the advanced residual prediction technique (advanced residual prediction,ARP) is a 3D video coding technique based on the residual correlation between viewpoints, which can effectively reduce the coding rate of non-basic viewpoints. The key point is to use the existing motion information to construct the residual prediction value for the current residuals. If the basic viewpoint corresponding block adopts the intra coding block, it does not contain motion information, which limits the use of advanced residual prediction to some extent. In order to solve the above problems, an improved advanced residual algorithm is proposed in this paper. In order to improve the prediction accuracy and improve the coding efficiency of ARP, the motion vectors of adjacent blocks of intra coding blocks are used to replace the zero vectors of intra coding blocks. The experimental results show that with the proposed algorithm, the coding rate of the non-basic viewpoint is reduced by 0.12 on average. For multi-view video, video with different viewpoints at the same time has strong correlation in image content. In 3D-HEVC, fine prediction unit partition algorithm is used to improve coding efficiency. In this paper, an inter-frame prediction unit optimization algorithm based on inter-view correlation is proposed to reduce the complexity of 3D-HEVC coding. According to the correlation between PU modes, the candidate list of current PU partitioning mode is constructed, and the PU candidate mode is simplified. In the transformation part of 3D-HEVC, TU adopts the residual quadtree transform structure, and obtains the best TU partition by traversing and comparing the rate-distortion cost of the transformation units of different depths. The residuals quadtree transform structure not only improves the coding efficiency, but also increases the computational complexity significantly. The advanced residual prediction technique proposed in 3D-HEVC makes full use of the correlation of the residual error and reduces the residual energy. Experimental statistics show that 97.38% of TU blocks using advanced residual prediction technology are not divided. Based on this, this paper proposes a preemptive TU partition optimization algorithm by defining the symbol bits using advanced residual prediction techniques. The experimental results show that when the coding efficiency and PSNR loss are very small, compared with the 3D-HEVC common test platform, the two optimization algorithms proposed in this paper reduce the average coding time of non-basic viewpoint texture images by 32.2b.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【相似文献】