基于四叉树划分的HEVC帧内编码快速算法研究

发布时间：2018-05-06 15:24

本文选题：HEVC + 四叉树划分　；参考：《北方工业大学》2017年硕士论文

【摘要】：高效视频编码标准HEVC(High Efficiency Video Coding)是新一代视频编解码标准,较上一代标准H.264在同等视频质量下能够节省一半的码流。在高质量视频所需的存储空间和带宽急剧增长的背景下,HEVC为视频的高效压缩提供了解决方案。但是HEVC在提高压缩比的同时,也引入了繁琐的四叉树递归结构,造成了编解码上的冗余,增加了编码时间。论文针对帧内预测的编码复杂度问题开展研究,在保证视频质量不发生明显下降的情况下,设计降低帧内编码复杂度的优化算法。主要研究成果如下:(1)提出了基于原始参考像素的四叉树划分提前终止算法。研究了基于重建参考像素和基于原始参考像素的帧内预测效果,发现基于后者能够提高预测准确度。算法以原始像素为参考样本,设计了四叉树划分的提前终止策略。通过比较基于大块及其四个子块的相应预测值,并结合左侧相邻块,提前判断当前块是否需要继续划分,设置了判断的最佳阈值。该算法有助于跳过一些不必要的率失真代价的计算过程,减少编码时间。实验结果表明,与参考代码相比平均可以节省25%的编码时间,且质量下降忽略不计。(2)提出了基于自适应SATD阈值的四叉树快速划分算法。哈达码变换的绝对误差和SATD值反映残差在频域上的大小,利用SATD能够对CU的大小做出判断。在每帧视频的四叉树递归划分过程中,选取前几个LCU(Large Code Unit)不同深度下的SATD值为参考,确定提前修剪和提前划分的SATD阈值,提前判断CU的大小;通过设定与量化参数QP相关的边界值,利用在前一步中无法提前判的CU的SATD值来自适应地更新这些阈值。实验结果表明,与参考代码相比,所提出的算法能够在仅增加0.9%的BD率的情况下,平均节省44%的编码时间。
[Abstract]:The efficient video coding standard HEVC(High Efficiency Video coding is a new generation video coding and decoding standard, which can save half of the code stream under the same video quality compared with the previous standard H.264. HEVC provides a solution for efficient video compression under the background of rapid increase of storage space and bandwidth required for high quality video. However, HEVC not only improves the compression ratio, but also introduces a complicated quadtree recursive structure, which results in redundancy in coding and decoding, and increases the coding time. In this paper, the coding complexity of intra prediction is studied, and an optimization algorithm to reduce the intra coding complexity is designed to ensure that the video quality does not decrease significantly. The main research results are as follows: (1) A quadtree partitioning early termination algorithm based on original reference pixels is proposed. The intra prediction results based on reconstructed reference pixels and original reference pixels are studied. It is found that the prediction accuracy can be improved based on the latter. The algorithm uses the original pixel as the reference sample and designs an early termination strategy for quadtree partitioning. By comparing the corresponding prediction values based on the block and its four sub-blocks and combining with the adjacent blocks on the left, this paper determines in advance whether the current block needs to be further divided, and sets the optimal threshold for the judgment. The algorithm can help to skip some unnecessary rate-distortion cost calculation process and reduce coding time. The experimental results show that compared with the reference code, the coding time can be saved by 25% on average, and the quality drop is ignored. (2) A fast quadtree partitioning algorithm based on adaptive SATD threshold is proposed. The absolute error and SATD value of Hada code transform reflect the magnitude of residual error in frequency domain, and the size of CU can be judged by SATD. In the process of quadtree recursive partition of each frame of video, the SATD values of the first several LCU(Large Code units at different depths are selected as reference to determine the SATD threshold of pruning and partitioning ahead of time, and to judge the size of CU in advance. By setting the boundary values associated with the quantization parameter QP, the SATD values of CU which could not be determined in advance in the previous step are used to update these thresholds adaptively. The experimental results show that compared with the reference code, the proposed algorithm can save an average coding time of 44% when the BD rate is only increased by 0.9%.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81

【相似文献】