视频编码标准HEVC低复杂度编码优化算法研究

发布时间：2018-05-23 16:31

本文选题：HEVC + 帧内预测　；参考：《华侨大学》2017年硕士论文

【摘要】：随着网络技术和多媒体技术的发展,视频应用的多样化趋势已经给视频压缩技术提出了更高的要求。为此,2010年4月VCEG和MPGE共同组建的JCT-VC,联合制定了新一代视频编码标准---H.265/HEVC,并与2013年正式发布标准。尽管H.265/HEVC比之前的H.264/AVC在相同的图像质量下,能节省50%的码率,但是对它的计算复杂度却提高了2到4倍。因此怎样提高新标准HEVC的编码速度,降低计算复杂度成为了现在研究者们的主要课题。本文也基于此目的,分别从帧间和帧内两个方面对HEVC进行编码优化,具体工作如下:在帧内编码方面,本文主要基于图像灰度直方图和空间相关性来实现的。因此整个优化过程分为两个部分,第一部分基于图像灰度直方图的灰度值分布概率进行LCU粗选,利用LCU块中的亮度值进行分类,先对LCU进行划分,划分成4个32×32,通过比较每一个32×32的灰度值得分布情况,若分布情况相似则不必要向下进行划分,也就是编码块可以保留32×32的大块,不必要进行更小的划分。第二部分是基于空间相关性对非LCU块进行编码块深度的决策,此方法利用帧内相邻单元之间存在很强的相关性理论对正在编码单元的划分进行参考,研究证明帧内编码单元块与它的上方,左方和左上方的单元块的划分相似度较高,因此,在对当前编码的CU块进行划分时,可以参考它相邻的CU块的划分方式。通过这两个步骤,帧内编码单元划分可以提前终止,节省了编码的时间,实验表明,此方法可以平均节省22.84%的编码时间,PSNR减少0.031dB,这说明视频的视觉质量没有下降太多,而压缩率也得到保障,但是编码时间却得到降低。在帧间编码方面,本文提出了在CU层编码内容相关性的优化算法,通过对阈值的设置提前预判编码单元CU的划分方式,从而可以有效降低编码时间。本文中提出一种基于图像内容相关性的帧间编码单元深度决策算法,从图像的亮度值出发,通过参考已编码帧之间的亮度均值的差值,通过设置阈值,提前终止CU单元的划分。为了减少计算量,本文只选取了每个LCU的21个关键像素点进行亮度均值差的计算进行比较,当某一块LCU的亮度均值与它前一帧的相同位置的LCU亮度均值差在阈值范围之内,则可以保留64×64的块划分方式,如果大于阈值而此时的参考块的划分为非64×64,则可以参考前一帧中相同位置的块划分,反之按照HM标准的划分进行遍历。实验表明,此算法可以平均节省28.959%编码时间,PSNR的变化也相对较小,只有0.0504dB。
[Abstract]:With the development of network technology and multimedia technology, the diversified trend of video applications has put forward higher requirements for video compression technology. Therefore, in April 2010, JCT-VC, which was set up by VCEG and MPGE, jointly developed a new generation of video coding standard-H.265 / HEVCand released the standard formally in 2013. Although H.265/HEVC can save 50% bit rate compared with previous H.264/AVC under the same image quality, its computational complexity is increased by 2 to 4 times. Therefore, how to improve the coding speed and reduce the computational complexity of the new standard HEVC has become a major issue for researchers. For this purpose, this paper also optimizes HEVC coding from two aspects of inter-frame and intra-frame. The main work is as follows: in intra-frame coding, this paper is mainly based on image gray histogram and spatial correlation. So the whole optimization process is divided into two parts. The first part is based on the gray value distribution probability of image grayscale histogram to carry on the LCU rough selection, using the brightness value in the LCU block to carry on the classification, first carries on the division to the LCU, It is divided into 4 32 脳 32. By comparing the value distribution of each 32 脳 32, if the distribution is similar, it is unnecessary to divide down, that is, the coding block can keep 32 脳 32 blocks, so it is unnecessary to divide them smaller. In the second part, based on spatial correlation, the coding depth of non-LCU blocks is determined. This method uses the theory of strong correlation between adjacent units in the frame to refer to the division of coding units. It is proved that the intra-coding unit block has a high similarity with its upper, upper left and upper left cell blocks. Therefore, when dividing the current coded CU blocks, we can refer to the partition methods of its adjacent CU blocks. Through these two steps, the intra-frame coding unit division can be terminated in advance, and the coding time is saved. Experiments show that this method can save an average coding time of 22.84% and PSNR decrease by 0.031 dB, which shows that the visual quality of video is not much reduced. The compression ratio is also guaranteed, but the coding time is reduced. In the aspect of inter-frame coding, this paper proposes an optimized algorithm of content correlation in CU layer. By setting the threshold, we can prejudge the division of CU in advance, so that the coding time can be reduced effectively. In this paper, a depth decision algorithm for interframe coding unit based on image content correlation is proposed. Starting from the brightness value of the image, by referring to the difference of the luminance mean between the encoded frames and setting the threshold value, the division of CU units is terminated ahead of time. In order to reduce the amount of computation, this paper only selects 21 key pixels of each LCU to calculate the luminance mean difference. When the luminance mean of a block of LCU is within the threshold range of the LCU luminance mean at the same position of the previous frame, the difference between the luminance mean of a block of LCU and that of the same position in the previous frame is within the threshold range. If the threshold is greater than the threshold and the reference block is not 64 脳 64, then the block partition at the same position in the previous frame can be referred to and traversed according to the HM standard. Experiments show that this algorithm can save an average of 28.959% coding time and the change of PSNR is relatively small, only 0.0504dB.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81

【相似文献】