3D-HEVC帧内编码优化算法研究

发布时间：2018-04-27 07:13

本文选题：3D-HEVC + JND　；参考：《北京交通大学》2017年硕士论文

【摘要】：近年来,摄像技术的飞速发展,带动了高清、超高清视频的飞速发展。虽然相应的硬件处理设备也有一定的改善,但视频压缩编码一直是一个重要的研究问题。相比过去的H.246编码标准,最新的国际视频编码标准HEVC在编码质量保持不变的情况下对高分辨率视频的压缩码率可以降低一半。高清视频发展到一定阶段,开始向立体视频发展。现在多视点裸眼三维以及360度全景视频的研究方兴未艾。全景视频能带给用户更好的沉浸式体验,但是其数据量远远大于高清视频。因此,在全景视频的编码过程中,HEVC的编码优势得以进一步的体现。本文重点研究基于HEVC标准的三维视频编码技术(3D-HEVC)。3D-HEVC相对于普通高清视频,编码的原理与过程更加复杂,编码的框架更加繁琐,编码所消耗的时间也大大增加。因此,针对上述问题,本文的工作主要包括:(1)人眼视觉系统具有各种掩藏效应及屏蔽效应,而且只能察觉超过一个特定阈值的信号变化,因此,本文将该视觉特性引入到3D-HEVC编码技术中,提出了基于像素域的恰可察失真(JustNoticeable Distortion,JND)模型的优化视频编码算法。该方法通过引入JND阈值的判断来代替复杂的率失真优化的过程,用于帧内编码单元(CodingUnit,CU)的快速划分,从而有效地降低了复杂度,节省了编码时间;(2)由于编码单元的总比特数(Total Bits)与视频本身的纹理复杂度密切相关,因此,本文所提出的帧内优化算法不仅考虑到了人眼的视觉特性,还进一步引入编码单元总比特数作为编码单元的划分的判定标准。本文根据这两个特征进行训练,寻找最合适的JND和总比特数阈值来优化编码单元的划分,从而可以省去大量的统计时间,直接确定相应的划分阈值,在保证编码重建质量的前提下,有效降低编码时间;(3)本文研究了全景图以及全景视频的生成原理,分析了全景图的柱面拼接算法。由于拼接后的全景图和全景视频数据量非常大,甚至于比高清视频还要大很多,目前使用H.264标准对其编码往往压缩效率极低。由于其数据结构的复杂性,目前国际上对全景图及全景视频编码尚未形成统一的标准。本文将HEVC标准算法应用到全景视频的编码中,提出了基于HEVC的全景视频编码算法,有效提高了编码的率失真性能。
[Abstract]:In recent years, the rapid development of camera technology, driving the rapid development of HD, ultra-HD video. Although the corresponding hardware processing equipment has been improved, video compression coding has been an important research problem. Compared with the previous H.246 coding standard, the latest international video coding standard, HEVC, can reduce the compression rate of high-resolution video by half while the coding quality remains the same. HD video develops to a certain stage and begins to develop to stereoscopic video. Now multi-view naked eye 3 D and 360-degree panoramic video are in the ascendant. Panoramic video can bring users a better immersive experience, but the amount of data is much larger than HD video. Therefore, in the panoramic video coding process, the advantages of HEVC coding can be further reflected. This paper focuses on the 3D video coding technology based on HEVC standard. 3D-HEVC.3D-HEVC is more complex in principle and process than in ordinary high-definition video, the frame of coding is more complicated, and the time consumed in coding is greatly increased. Therefore, in response to the above problems, the work of this paper mainly includes: 1) the human visual system has a variety of hiding effects and shielding effects, and can only detect signal changes over a specific threshold. In this paper, the visual characteristic is introduced into the 3D-HEVC coding technology, and an optimized video coding algorithm based on the pixel domain is proposed. In this method, the decision of JND threshold is introduced to replace the complex rate-distortion optimization process, and it is used for fast partition of coding unit within frames, thus reducing the complexity effectively. Since the total bit number of the coding unit is closely related to the texture complexity of the video itself, the intra-frame optimization algorithm proposed in this paper not only takes into account the visual characteristics of the human eye, but also reduces the coding time. Furthermore, the total bit number of the coding unit is introduced as the criterion for the partition of the coding unit. According to these two characteristics, this paper trains to find the most suitable JND and total bit number threshold to optimize the division of coding units, thus can save a lot of statistical time, and directly determine the corresponding threshold. In this paper, the principle of panoramic image and panoramic video generation is studied, and the cylindrical mosaic algorithm of panoramic image is analyzed. Because of the large amount of panoramic image and panoramic video after stitching, even more than high-definition video, the coding efficiency of H.264 standard is very low. Due to the complexity of its data structure, there is no uniform standard for panorama and panoramic video coding in the world. In this paper, the HEVC standard algorithm is applied to the panoramic video coding, and a panoramic video coding algorithm based on HEVC is proposed, which can effectively improve the rate distortion performance of the coding.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81

【参考文献】