视频压缩中的高效预测技术研究
发布时间：2018-08-14 13:53
【摘要】：视频如今已经走入每个人的日常生活中,成为不可缺少的一部分,并且随着人们感官要求越来越高,高清视频、3D视频逐渐成为市场主流,然而高质量的视频也需要更大的存储空间和传输带宽。近年来国际通用压缩标准MPEG-4、H.264、HEVC等,利用视频图像的相关性去除其中的冗余信息,我国自主研发的第二代音视频标准AVS2几乎与国际标准HEVC压缩效果相当,同时3D-AVS标准于2014年完成,供大家对3D视频的压缩进行研究。本文就当前存在的传统视频和多视点视频技术分别做了研究。在传统视频方面,针对国际视频编码标准HEVC的帧内预测部分,提出两个帧内预测算法;在多视点视频方面,针对国内的3D视频编码标准3D-AVS,提出了一个关于视差矢量的获取方案,主要工作分为以下三个部分。第一部分提出了基于马尔科夫模型梯度自适应帧内预测算法,该算法将马尔科夫模型与梯度自适应算法相结合,提升了帧内预测部分的性能。该算法在HEVC参考软件HM16.0基础上实现,实验结果显示,亮度分量的平均BDrate减少1.0%,两个色度分量的平均BD-rate减少1.2%和1.3%。第二部分提出了基于PDPC(Position dependent intra prediction combination)的梯度自适应帧内预测算法,参考了视频编码联合工作组正在研究的下一代视频编码技术中的PDPC帧内预测算法,在此基础上,充分利用了当前像素与周围像素的相关性。实验在HEVC参考软件JEM-2.0平台上实现,相比于JEM-2.0,亮度分量的平均BD-rate减少0.2%,两个色度分量的平均BD-rate减少0.4%。第三部分针对多视点视频,提出了基于时域和空域的局部视差矢量获取算法,该算法在原有全局视差矢量获取算法和局部视差矢量获取算法的基础上,同时考虑了时域参考帧和当前帧的局部自适应相邻区域的视差矢量。该算法在3D-AVS参考软件RFD8.0平台上实现,实验结果显示,相比于RFD8.0,在LDP配置下合成视点的平均BD-rate减少约0.12%。本文提出的三个算法都取得了较好的实验结果,相比于原有视频编码标准和现阶段该领域的相关算法都有一定的性能增益。
[Abstract]:Nowadays, video has become an indispensable part of everyone's daily life, and as people's senses become more and more demanding, high-definition video / 3D video has gradually become the mainstream of the market. However, high-quality video also needs more storage space and transmission bandwidth. In recent years, the international general compression standard MPEG-4 / H.264 / HEVC and so on, using the correlation of video image to remove the redundant information, the second generation audio and video standard AVS2 developed by our country is almost equal to the international standard HEVC compression effect, and the 3D-AVS standard was completed in 2014. For everyone to study the compression of 3D video. In this paper, the existing traditional video and multi-view video technology are studied. In the traditional video aspect, two intra prediction algorithms are proposed for the intra prediction part of the international video coding standard (HEVC), and a scheme for acquiring parallax vectors is proposed for the 3D video coding standard 3D-AVS, which is a domestic standard of 3D video coding. The main work is divided into the following three parts. In the first part, a gradient adaptive intra prediction algorithm based on Markov model is proposed, which combines Markov model with gradient adaptive algorithm to improve the performance of intra prediction. The algorithm is implemented on the basis of HEVC reference software HM16.0. The experimental results show that the average BDrate of the luminance component is reduced by 1.0%, the average BD-rate of the two chrominance components is reduced by 1.2% and 1.3%. In the second part, a gradient adaptive intra prediction algorithm based on PDPC (Position dependent intra prediction combination) is proposed, which refers to the PDPC intra prediction algorithm in the next generation video coding technology which is being studied by the Joint Video coding working Group. Make full use of the correlation between the current pixel and the surrounding pixel. The experiment is implemented on the HEVC reference software JEM-2.0 platform. Compared with JEM-2.0, the average BD-rate of the luminance component is 0.2 and the average BD-rate of the two chrominance components is 0.4. In the third part, a local parallax vector acquisition algorithm based on time domain and spatial domain is proposed for multi-view video. The algorithm is based on the original global parallax vector acquisition algorithm and local parallax vector acquisition algorithm. The parallax vectors of the local adaptive adjacent regions of the time domain reference frame and the current frame are also considered. The algorithm is implemented on the 3D-AVS reference software RFD8.0 platform. The experimental results show that the average BD-rate of the composite viewpoint in LDP configuration is about 0.12% less than that of RFD8.0. The three algorithms presented in this paper have achieved good experimental results, and have a certain performance gain compared with the original video coding standards and the current related algorithms in this field.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2183069
[Abstract]:Nowadays, video has become an indispensable part of everyone's daily life, and as people's senses become more and more demanding, high-definition video / 3D video has gradually become the mainstream of the market. However, high-quality video also needs more storage space and transmission bandwidth. In recent years, the international general compression standard MPEG-4 / H.264 / HEVC and so on, using the correlation of video image to remove the redundant information, the second generation audio and video standard AVS2 developed by our country is almost equal to the international standard HEVC compression effect, and the 3D-AVS standard was completed in 2014. For everyone to study the compression of 3D video. In this paper, the existing traditional video and multi-view video technology are studied. In the traditional video aspect, two intra prediction algorithms are proposed for the intra prediction part of the international video coding standard (HEVC), and a scheme for acquiring parallax vectors is proposed for the 3D video coding standard 3D-AVS, which is a domestic standard of 3D video coding. The main work is divided into the following three parts. In the first part, a gradient adaptive intra prediction algorithm based on Markov model is proposed, which combines Markov model with gradient adaptive algorithm to improve the performance of intra prediction. The algorithm is implemented on the basis of HEVC reference software HM16.0. The experimental results show that the average BDrate of the luminance component is reduced by 1.0%, the average BD-rate of the two chrominance components is reduced by 1.2% and 1.3%. In the second part, a gradient adaptive intra prediction algorithm based on PDPC (Position dependent intra prediction combination) is proposed, which refers to the PDPC intra prediction algorithm in the next generation video coding technology which is being studied by the Joint Video coding working Group. Make full use of the correlation between the current pixel and the surrounding pixel. The experiment is implemented on the HEVC reference software JEM-2.0 platform. Compared with JEM-2.0, the average BD-rate of the luminance component is 0.2 and the average BD-rate of the two chrominance components is 0.4. In the third part, a local parallax vector acquisition algorithm based on time domain and spatial domain is proposed for multi-view video. The algorithm is based on the original global parallax vector acquisition algorithm and local parallax vector acquisition algorithm. The parallax vectors of the local adaptive adjacent regions of the time domain reference frame and the current frame are also considered. The algorithm is implemented on the 3D-AVS reference software RFD8.0 platform. The experimental results show that the average BD-rate of the composite viewpoint in LDP configuration is about 0.12% less than that of RFD8.0. The three algorithms presented in this paper have achieved good experimental results, and have a certain performance gain compared with the original video coding standards and the current related algorithms in this field.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前5条
1 黄铁军;;AVS2标准及未来展望[J];电视技术;2014年22期
2 ;AVS 3D Video Coding Technology and System[J];ZTE Communications;2012年02期
3 ;Introduction to the High-Efficiency Video Coding Standard[J];ZTE Communications;2012年02期
4 杨升;张德鹏;;高清电视发展的国际国内环境分析[J];广播与电视技术;2009年05期
5 万帅,王新岱;新一代视频压缩标准H.264[J];中国有线电视;2003年18期
，本文编号：2183069
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2183069.html
