基于纹理特性与视觉关注度的HEVC优化研究

发布时间：2019-05-31 18:00

【摘要】：随着网络的发展与视频应用的普及,用户对视频质量的要求越来越高,高质量的视频需要大量数据描述画面细节,导致视频数据量激增。高性能视频编码(High Efficiency Video Coding,HEVC)是面向高分辨率视频的新一代编码标准,其核心目标是在H.264/AVC High Profile的基础上,将视频压缩效率提高一倍。但压缩效率提高的同时也带来了较高的计算复杂度与较长的编码时间,严重影响了HEVC的推广与应用。在视频中,物体的纹理通过局部区域像素的排列与变化来表现,通常呈缓慢或周期性变化,具有一定的规律性。HEVC以编码单元(Coding Unit,CU)为基本单位对图像进行编码,将纹理简单区域划分为低深度级别的大尺寸CU,将纹理复杂区域划分为高深度级别的小尺寸CU,对纹理相似的区域CU深度划分相近。但CU划分算法计算复杂度高,成为制约HEVC性能的主要因素之一。所以,在HEVC中考虑视频的纹理特性可以预测CU划分深度,降低编码计算复杂度,有效减少编码时间。另一方面,眼睛是各种视频信号的最终受体,视频质量也可以说是人眼对视频感知的主观质量。人类视觉系统并非平等地关注视频中所有区域,在视频编码中根据视觉对图像区域关注度的不同调整码率资源分配,可有效去除视觉冗余,提升压缩性能。因此,基于纹理特性与视觉关注度的HEVC优化研究能够有效提高HEVC编码性能,具有重要的理论意义和广阔的应用价值。首先在深入研究CU划分原理的基础上,提出一种基于Canny算子的CU快速划分算法,使CU提前进入子划分,降低编码复杂度,加快编码过程。然后根据人眼感知特性建立视觉关注度模型,计算当前最大编码单元(Largest Coding Unit,LCU)关注度,调节不同关注度区域的码率资源分配,实现自适应编码压缩,提高整体压缩比。本文的主要研究内容包括以下三个方面:(1)研究CU初始深度预测算法,优化CU划分。首先研究CU划分深度与其邻域及参考帧相同位置CU深度的相关性,推导出CU初始深度与纹理分布的数学关系;然后利用Canny分割算子边缘定位精度高、连续性良好等优点,分割关键帧的纹理区域,并判断纹理在当前CU与邻域中分布关系;最后根据纹理分布情况预测CU初始深度,简化CU划分递归过程,降低编码复杂度,加快编码过程。(2)模拟人类视觉系统的选择性注意机制建立关注度模型。根据视觉感知特性引入运动性因子、纹理复杂度因子、对比度因子与亮度因子建立视觉关注度模型。为保证编码效率,采用计算复杂度低、鲁棒性强的灰度投影法计算运动性因子,基于亮度分布情况计算纹理复杂度因子,采用像素四近邻算法计算对比度因子,采用编码单元四近邻算法计算亮度因子。(3)根据CU关注度的不同,调整码率资源分配,实现自适应编码压缩。根据人眼更加关注结构性失真而非像素点失真的特点,对高关注度LCU使用构建的结构相似性失真优化算法而非传统的误差平方和算法,对低关注度LCU利用关注度修正拉格朗日因子,实现对低关注度区域粗量化,达到提高压缩比,减少码率的效果。
[Abstract]:With the development of the network and the popularization of the video application, the demand for the video quality of the user is higher and higher, and the high-quality video needs a large amount of data to describe the detail of the picture, resulting in a sharp increase in the amount of video data. High-performance Video Coding (HEVC) is a new-generation coding standard for high-resolution video. Its core goal is to double the video compression efficiency on the basis of the H.264/ AVC High Profile. But the compression efficiency is improved, higher calculation complexity and long coding time are also brought, and the popularization and application of the HEVC are seriously affected. In video, the texture of an object is represented by the arrangement and variation of the local area pixels, usually in a slow or periodic manner, with a certain regularity. HEVC is used to encode an image by a coding unit (CU), and the texture simple area is divided into a large-size CU with a low depth level, and the texture complex area is divided into a small-size CU with a high depth level, and the depth of the area CU with similar texture is similar. However, the calculation complexity of the CU is high and becomes one of the main factors that restrict the performance of the HEVC. Therefore, considering the texture characteristics of the video in the HEVC, the division depth of the CU can be predicted, the coding calculation complexity is reduced, and the coding time is effectively reduced. On the other hand, the eye is the final receptor of various video signals, and the video quality can also be said to be the subjective quality of the human eye's perception of the video. The human vision system is not equally concerned with all the areas in the video, and can effectively remove the visual redundancy and improve the compression performance according to the different adjustment code rate resource allocation of the attention of the visual on the image area in the video coding. Therefore, the HEVC optimization research based on the texture characteristic and the visual attention can effectively improve the HEVC coding performance, and has important theoretical significance and wide application value. First, on the basis of in-depth study of the principle of CU division, a fast algorithm of CU based on Canny operator is proposed, which makes the CU enter the sub-division in advance, reduce the coding complexity and speed up the coding process. Then the visual attention model is established according to the perception characteristic of the human eye, the attention of the current maximum coding unit (LCU) is calculated, the code rate resource allocation of the different attention regions is adjusted, the adaptive coding compression is realized, and the overall compression ratio is improved. The main research contents of this paper include the following three aspects: (1) study the initial depth prediction algorithm of the CU, and optimize the CU division. Firstly, the relationship between the depth of the CU division depth and its neighborhood and the same position of the reference frame is studied, the mathematical relation between the initial depth of the CU and the texture distribution is derived, and then the texture region of the key frame is divided by using the advantages of high edge positioning accuracy and good continuity of the Canny division operator. And finally, the initial depth of the CU is predicted according to the texture distribution condition, a recursive process of the CU is simplified, the coding complexity is reduced, and the coding process is accelerated. And (2) simulating the selective attention mechanism of the human vision system to establish the attention model. According to the visual perception characteristics, a visual attention model is established by introducing a motility factor, a texture complexity factor, a contrast factor and a brightness factor. In order to guarantee the coding efficiency, the motion factor is calculated by the gray-scale projection method with low computational complexity and strong robustness, the texture complexity factor is calculated based on the brightness distribution, the contrast factor is calculated by using the four-neighbor algorithm of the pixel, and the brightness factor is calculated by adopting the four-neighbor algorithm of the coding unit. And (3) adjusting the code rate resource allocation according to the different degree of the CU, so as to realize the self-adaptive coding compression. according to the characteristic that the human eye is more concerned with the structural distortion and the non-pixel point distortion, the structure similarity distortion optimization algorithm constructed by the high-degree-of-interest LCU is used instead of the non-traditional error sum-of-square algorithm, and the Lagrange factor is corrected for the low-degree-of-attention LCU by the degree of attention, The coarse quantization of the low-degree-of-attention area is realized, and the effect of improving the compression ratio and reducing the code rate is achieved.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81

【参考文献】