多视点视频编码的研究现状及其展望

阐述了多视点视频编码(MVC)的主要研究问题.其中首先介绍了MVC的体系结构和发展过程;然后详细讨论了MVC的研究内容,包括预测结构、提高MVC编码效率的技术和高层语法;最后在总结MVC研究现状的基础上,提出了MVC在3D视频应用中的研究思路.     

一种优化的多视点视频编码预测结构

首先分析了JMVC的参考预测结构,利用1个Go P(Group of Picture)中不同时间层上的帧与其参考帧之间的相关度不同和不同时间层上的帧数目也不同的特点,提出了去除奇数视点最高和次高时间层的视点间预测,并增加偶数视点非关键帧的最低和次低时间层的视点间预测的预测结构。实验结果表明,提出的预测结构基本不影响视频质量,且表现出了更加出色的编码效率,编码的复杂度减少幅度高达15%,输出码率也有所减少,较好地改善了编码的实时性能。     

采用率失真与模式特征的多视点视频编码快速模式选择

现有的多视点视频编码使用了分层B帧（Hierarchical B Picture, HBP）的预测结构,其帧内预测、帧间预测以及视点间预测的模式选择给多视点视频编码带来了庞大的计算复杂度。针对这一问题,我们在分析了JMVC模式分布比例的基础上,提出了一个快速帧间模式选择的算法。这种算法利用率失真代价和预测模式特征之间的关系来及时判定最优模式:如果上一尺寸预测模式的率失真代价小于当前尺寸预测模式的率失真代价则认为上一预测模式为最优模式,跳过检查其他更小尺寸的预测模式;反之,如果上一尺寸的预测模式的率失真代价大于当前尺寸的预测模式的率失真代价,则继续检查其他更小的尺寸。这样,通过提前终止一些不必要的模式选择过程,多视点视频编码的计算量得到大幅的降低。实验结果表明:所提算法能在保持JMVC中全搜索算法的编码效率同时,使计算复杂度减少了81.66%。      

面向三维高效视频编码的深度图错误隐藏

基于多视点视频序列视点内、视点间存在的相关性,并结合视点间运动矢量共享技术,该文提出一种面向3维高效视频编码中深度序列传输丢包的错误隐藏算法。首先,根据3D高效视频编码(3D-HEVC)的分层B帧预测(HBP)结构和深度图纹理特征,将深度图丢失块分成运动块和静止块;然后,对于受损运动块,使用结合纹理结构的外边界匹配准则来选择相对最优的运动/视差矢量进行基于位移矢量补偿的错误掩盖,而对受损静止块采用参考帧直接拷贝进行快速错误隐藏;最后,使用参考帧拆分重组来获取新的运动/视差补偿块对修复质量较差的重建块进行质量提升。实验结果表明:相较于近年提出的对比算法,该文算法隐藏后的深度帧平均峰值信噪比(PSNR)能提升0.25～2.03 dB,结构相似度测量值(SSIM)能提升0.001～0.006,且修复区域的主观视觉质量与原始深度图更接近。     

一种改进的多视点视频编码预测结构

多视点视频编码(Muti-view Video Coding,MVC)标准采用预测关系固定的编码结构,不能根据视频序列调整预测关系,导致无法获得更好的编码效率.本文针对这一缺陷提出一种确定预测结构的方法,首先对图像进行下采样获得粗粒度图像,然后利用粗粒度图像间的灰度方差获得灰度相近的视点,最后对所选取的灰度相近的视点运用全局视差确定I-视点.经实验验证,本文的方法可以提高编码效率并减少编码中的缓存容量.     

面向HBP编码格式的立体视频B帧整帧丢失分层错误隐藏算法

针对分层B帧(HBP)编码格式的立体视频B帧整帧丢失的问题,该文分析了双视点视频中存在的视点间运动矢量相关性,提出一种分层错误隐藏算法。该算法与当前主流的方法有两大不同:一是该算法采用分级隐藏,根据B帧的重要性等级不同采用不同的错误隐藏方法;二是该算法考虑了相邻视点序列之间的宏块运动矢量相关性。实验表明,该算法的性能优于当前常用的H.264多视点视频整帧丢失错误隐藏方法。     

一种基于多视点视频的低复杂度自适应环路滤波算法

自适应环路滤波(ALF)是面向高清的多视点视频编码(MVC)中的一项新的滤波技术,虽然能提高视频压缩的主观质量和客观质量,但是其复杂度过高阻碍了其实时应用。本文提出了一种低复杂度的ALF算法,充分利用了视点间相关性和层间相关性以减少亮度和色度分量的分割次数,并且能自适应跳过大部分帧的块控制(DBC)处理过程。实验结果表明,所提出的算法能在保证视频主观质量和客观质量基本不变的情况下减少61%左右的ALF处理时间。因此,所提出的方法在编码效率和复杂度上做到了很好的折中处理,并且能很好地改善ALF的性能。     

基于转移概率的多视点视频快速模式选择算法

针对多视点视频编码(MVC)中部分大宏块模式(SKI P/DIRECT模式,帧间16×16模式)相对于其它模式的计算复杂 度低,并且在编码所确定的最优模式中占有相当高比重的特点,本文提出一种基于转移概率 的多视点视频 快速模式选择算法。首先根据多视点视频当前编码宏块对应时间和视点间宏块及周围宏块的 位置几何关系, 建立宏块模式参考模型;其次比较参考模型中模式的时间相似度和视点间相似度;最后计算 模式之间的转 移概率,确定提前结束大宏块模式选择过程的阈值。实验结果表明,本文所提出的快速算法 平均节约MVC时间80.93%,编码质量平均下降0.04dB,码率平均增加0.27%。     

一种新的多视点视频编码预测结构

多视点视频编码的参考预测结构虽然具有高效的压缩效率,但是随机访问性能较差.在综合考虑了编码效率和随机访问等因素的基础上,提出了一种改进的多视点视频预测结构编码方法.首先,分析视点与视点间的相关性来调整Ⅰ视点的位置,然后在此基础上有效地增加B视点的数目,以获得较好的综合性能.实验结果表明,该方法与传统的参考预测结构编码方法相比,提高了编码效率,并且具有较好的随机访问性能.     

基于感知的多视点视频编码宏块模式选择快速算法

多视点视频编码(MVC)采取可变块模式选择技术和 多参考帧技术显著提高了编码的压缩效率,但同时带来了巨大的 编码计算复杂度。为了降低MVC的计算复杂度,提出基于感知的快速MVC宏块模式选择 算法。基于人眼视觉感知的特点,利用视觉恰可察觉失真(JND)的概念建立MVC宏块的最优模 式和JND的联系,并利用该联系确定早期结束最优宏块模式选择过程的阈值,根据当前编码 宏块的JND与阈值的 关系自适应地减少每个编码宏块的模式搜索次数,进而减少MVC的方向搜索和参考帧搜索的 次数,以降 低编码的复杂度,提高MVC速度。实验结果显示,对于不同运动特性、内容、纹理信息、相 机间距和图像 尺寸的测试序列,提出的快速算法在率失真性能几乎不变的情况下平均节约76.00% 编码时间。     

一种新的视点间预测结构

多视点视频编码除应具有较高的编码效率外,还应该包括后向兼容性、时间随机访问和视点可分级性等,这些都主要取决于所采用的预测结构。目前所提供的多视点视频编码(Joint Multi-view Video Coding, JMVC)采用固定的视点间预测结构,难以适应复杂情况的多视点视频编码。该文综合考虑编码效率和用户随机访问等因素,根据多视点视频相关性分析自适应调整视点间预测结构,以获得较好的编码综合性能。试验结果表明,与JMVC相比,该文的方法在提高编码效率的同时,有较好的随机访问性能。     

应用于多视点视频压缩的多参考B帧快速编码算法

基于H.264平台就多视点视频压缩中多参考帧技术存在计算复杂度高的现状,根据多视点视频序列特性,提出了一种针对多参考B帧的快速多视点视频编码压缩算法。新算法改进了多参考帧列表策略,并结合多视点视频序列的时空域相关性特性,提出了适用于多视点视频编码的参考帧列表调整策略。此外,还针对多种块模式编码引入了块模式预判策略。新算法在保证率失真性能的基础上,有效地降低了计算复杂度,提高了多视点视频编码速度。虽然本文是基于M-Picture编码结构实现了新算法,但它同样适用于其它应用多参考B帧预测技术的多视点视频编码结构中。     

Multi-view video coding with view interpolation prediction for 2D camera arrays

An efficient compression algorithm for multi-view video sequences, which are captured by two-dimensional (2D) camera arrays, is proposed in this work. First, we propose a novel prediction structure, called three-dimensional hierarchical B prediction (3DHBP), which can efficiently reduce horizontal inter-view redundancies, vertical inter-view redundancies, and temporal redundancies in multi-view videos. Second, we develop a view interpolation scheme based on the bilateral disparity estimation. The interpolation scheme yields high quality view frames by adapting disparity estimation and compensation procedures using the information in neighboring frames. Simulation results demonstrate that the proposed multi-view video coding algorithm provides significantly better rate–distortion (R–D) performance than the conventional algorithm, by employing the 3DHBP structure and using interpolated view frames as additional reference frames.     

SIFT-flow-based color correction for multi-view video

During the multi-view video acquisition, color variation across the views tends to be incurred due to different camera positions, orientations, and local lighting conditions. Such color variation will inevitably deteriorate the performance of the follow-up multi-view video processing, such as multi-view video coding (MVC). To address this problem, an effective color correction algorithm, called the SIFT flow-based color correction (SFCC), is proposed in this paper. First, the SIFT-flow technique is used to establish point-to-point correspondences across all the views of the multi-view video. The average color is then computed based on those identified common corresponding points and used as the reference color. By minimizing the energy of the difference yielded between the color of those identified common corresponding points in each view with respect to the reference color, the color correction matrix for each view can be obtained and used to correct its color. Experimental results have shown that the proposed SFCC algorithm is able to effectively eliminate the color variation inherited in multi-view video. By further exploiting the developed SFCC algorithm as a pre-processing for the MVC, extensive simulation results have shown that the coding efficiency of the color-corrected multi-view video can be greatly improved (on average, 0.85 dB, 1.27 dB and 1.63 dB gain for Y, U, and V components, respectively), compared with that of the original multi-view video without color correction.     

多视点视频编码中图像分组技术的研究

在多视点视频编码中,引入了视点间运动补偿预测技术,而在视点间预测过程中,由于视差矢量的使用,使得GoP长度对编码效率的影响与一般H.264/AVC编码有所不同。在分析了可分级B帧的预测结构对多视点视频编码性能的影响基础上,通过对不同视频序列设置不同GoP的大小,经过多次反复实验,而后对实验结果进行分析,找到了在一定数值范围内的GoP。通过该GoP得到了较好的PSNR和模值较小的MV或者DV,节约了一定的编码时间。     

A motion vector prediction method for multi-view video coding

This paper proposes a new motion vector (MV) prediction method in multi-view video coding (MVC). In order to exploit the information in adjacent views, inter-view MVs as well as temporal MVs are used in conventional MVC. Since the inter-view MVs are usually uncorrelated with the temporal MVs and most neighboring partitions have temporal MVs only, the conventional DPCM coding gain of inter-view MV is very low and thus the inter-view MVs are seldom selected. In order to increase the probability of inter-view MV selection, we define a virtual inter-view MV which can be generated from temporal MVs. Then, an inter-view MV is predicted using these neighboring virtual inter-view MVs, leading to less prediction error than using the temporal MVs. As a result, bit-rates are decreased by up to 9% for the view-temporal prediction structure.     

Efficient disparity vector prediction schemes with modified P frame for 2D camera arrays

An efficient disparity estimation algorithm for multi-view video sequences, recorded by a two-dimensional camera array in which the cameras are spaced equidistantly, is presented. Because of the strong geometrical relationship among views, the disparity vectors of a certain view can for most blocks be derived from the disparity vectors of other views. A frame constructed using that idea is called a D frame in this work. Three new prediction schemes which contain D frames are proposed for encoding 5 × 3 multi-view video sequences. The schemes are applied to several multi-view image sequences taken from a camera-array and they are compared in terms of quality, bit-rate and complexity. The experimental results show that the proposed prediction schemes significantly decrease the complexity of the encoder at a very low cost of quality and/or bit-rate.     

Shareability-Exclusivity Representation on Product Grassmann Manifolds for Multi-camera video clustering

With the rapid popularity of multi-camera networks, one human action is usually captured by multiple cameras located at different angles simultaneously. Multi-camera videos contain the distinct perspectives of one action, therefore multiple views can overcome the impacts of illumination and occlusion. In this paper, we propose a novel multi-camera video clustering model, named Shareability-Exclusivity Representation on Product Grassmann Manifolds (PGM-SER), to address two key issues in traditional multi-view clustering methods (MVC): (1) Most MVC methods directly construct a shared similarity matrix by fusing multi-view data or their corresponding similarity matrices, which ignores the exclusive information in each view; (2) Most MVC methods are designed for multi-view vectorial data, which cannot handle the nonlinear manifold structure hidden in multi-camera videos. The proposed PGM-SER firstly adopts Product Grassmann Manifolds to represent multi-camera videos, then simultaneously learn their shared and exclusive information in global structures to achieve multi-camera video clustering. We provide an effective optimization algorithm to solve PGM-SER and present the corresponding convergence analysis. Finally, PGM-SER is tested on three multi-camera human action video datasets and obtain satisfied experimental results.