节点文献

AVS视频编码中帧间预测算法及全零块检测研究

Research on Inter Prediction Algorithm and All-zero Block Detection of AVS Video Coding

【作者】 陈栋

【导师】 张卫宁;

【作者基本信息】 山东大学 , 信号与信息处理, 2010, 硕士

【摘要】 随着人们对多媒体信息需求的日益增长,多媒体技术的发展突飞猛进。作为多媒体技术的核心,视频处理技术也得到了高速发展。大量的视频数据对传输带宽和储存空间都提出了很高的要求,所以对视频数据进行压缩化的视频编解码技术已经成为国内外研究和工业应用的热点之一。AVS (Advanced Audio-Video Coding-Decoding Standard)是我国“数字音视频编解码技术标准工作组”自主研发的第二代音视频编码标准,是当前世界上最先进的音视频编码标准之一,并且具有我国自主的知识产权。AVS具有复杂度低,压缩比高,图像质量高,应用范围广,专利费低等优点,代表着我国数字音视频产业的技术水平及发展方向。本文对AVS编码器的核心算法进行了改进,并进行了代码实现,显著提高了编码器的编码性能。本文介绍了AVS视频编解码标准的编码原理,编码流程及其关键技术,包括:帧间预测、帧内预测、亚像素差值、整数变换与量化、环路滤波、熵编码等等。将AVS视频编解码标准与国外主流标准H.264/AVC进行对比,从技术水平与实际应用两方面说明了AVS标准在国内音视频产业中的发展潜力。之后,以编解码技术标准工作组提供的编码器参考模型rm52j为参考模型,进行了模块结构和代码实现的整体介绍,从而为接下来的算法改进与实现建立坚实的理论知识基础。在参考模型中,帧间预测所采用的是传统的率失真优化模式选择算法,需要遍历所有帧间预测模式,所以会占用大量的时间与系统资源。本文提出了一种将时域与纹理相结合的快速帧间预测模式选择算法,结合了原始视频图像序列中图像之间的时域相关性,跳过宏块预判和宏块自身的纹理特性,从多个方面实现了对模式选择算法的优化。对于背景相对固定的视频序列或者运动相对激烈的视频序列都有明显的速度提升,在保证视频本身编码性能的前提下,显著节省了编码时间。根据AVS标准中整数变换的特点,本文采用了一种基于8×8离散整数余弦变换的一种有效的全零块判别准则,并推导出了相应的全零块检测门限。全零块检测是面向低比特率的视频编码器常用的优化方法之一,特别是与基于跳过宏块预判的帧间模式预测相结合,可以省去变换、量化、反变换、反量化、块重建等所带来的大量计算。在保证视频图像质量和比特率几乎不变的情况下,大幅度节省编码时间。将全零块的检测算法应用到运动估计阶段,能够减少运动搜索的次数,提高运动估计效率,增强编码器的性能。

【Abstract】 For the increasing demand of multimedia information, the multimedia technology is developing very quickly. As the core of multimedia technology, video processing technology is quick developing too. A large number of video data has put forward a very high demand of transmission bandwidth and storage space, so the video codec technology which encodes and decodes the video sequence has become one of the hot issues of the research and industrial application. AVS (Advanced Audio-Video Coding-Decoding Standard) video standard is developed by the Audio Video Coding Standard Working Group of China, which is one of the most advanced video compression standards available and has the own intellectual property rights. AVS has low complexity, high compression ratio, high image quality, wide application range, low patent cost. This paper carries out AVS video encode optimization in aspects of structural adjustment, source code optimization and optimized implementation of core algorithm. These lead to obvious raise in encoding efficiency.An introduction is given which is about the encode principle and encode flow of AVS video codec standard, together with the key technologies, including inter prediction, intra prediction, sub-pixel interpolation, transform and quantization, loop filter and entropy coding, etc. The comparison between AVS and H.264/AVC which is the most popular video codec standard abroad is given. Then, based on the reference software rm52j provided by AVS video codec technology standard group, we do the structural and source code introduction, which will provide foundation for the following algorithm improvement and code implement.Rate distortion optimization mode decision is used in AVS standard. All the modes of inter prediction will be traversed, which results in severely time-consuming. A fast inter prediction mode decision algorithm based on the time and texture characteristic of macroblocks. Speed improvement is very obvious for both video with fixed background and fast-motion video. This algorithm can save the coding time obviously while maintaining the coding performance.According to the characteristic of integral transform, this paper uses an efficient all-zero block determination rule for 8×8 integral discrete Cosine transform and the corresponding threshold. All-zero detection is one of the common optimization methods oriented towards low bit rate video coding, especially when combined with inter mode prediction based on skip blocks prediction. By using the algorithm the time would be reduced greatly which taken by implementing transform, quantization, inverse transform, inverse quantization and block reconstruction with negligible loss of PSNR and increment of bit rate. Using all zero block detection algorithm in motion estimation will reduce the times of motion search, improve motion estimation efficiency and enhance the encode performance.

  • 【网络出版投稿人】 山东大学
  • 【网络出版年期】2010年 08期
节点文献中: