三维视频主客观质量评价方法与感知优化编码研究

发布时间：2019-02-09 18:59

【摘要】：目前对于普通家庭用户而言,在线观看三维视频的效果无法令人满意,不仅画面模糊而且三维立体感较差。这一方面是因为三维视频的数据量巨大,过度的压缩对三维视频的立体感与深度感体验影响很大。另一方面,如何对三维视频的质量进行合理的评价是一个还没有被很好地解决的问题。三维视频的失真类型与传统的二维视频存在区别,直接套用己有的视频质量评价方法效果不佳,且三维视频的用户质量体验还包含了三维视频深度感质量等更复杂的影响因素。深度图作为一种三维视频辅助信息目前正在得到越来越广泛的应用。对于基于深度图的三维视频系统而言,虚拟视点的视频图像质量好坏将会影响整个系统的用户体验质量。第二章主要对存在纹理/深度压缩失真的虚拟绘制视点视频的主观和客观质量评价进行了研究。主观数据集的设计原则是确保用于测试的虚拟绘制视频的质量覆盖范围足够广并且彼此之间具有一定的质量区分度。为此,每个纹理/深度视频压缩量化参数组合都是从大量的候选中进行仔细挑选的。所建立的虚拟绘制视点视频主观质量评价数据集目前己被众多国内外的知名研究机构所使用。另外,还提出了一种全参考的虚拟绘制视点视频客观质量评价算法。该算法主要关注由深度图压缩以及视点绘制过程本身带来的时域闪烁失真。实验结果表明,所提出的算法在完整数据集上的性能要显著优于现有的客观视频质量评价算法,在时域闪烁失真较为明显的子集上相对于其它算法的优势更加明显。第三章重点研究了三维视频的深度感质量评价,通过主观实验对图像失真如何影响三维立体视频的深度感知质量进行了深入探索,实验数据集既包括了对称失真的情况,又包括了非对称失真的情况。主观实验的结果证明图像细节的丢失会影响深度感知。主观实验得到的三维立体视频深度感知质量分数以及图像质量分数已经对外公开发布。同时,该章还提出了一种客观评价算法用于测量图像失真引起的深度感知质量下降。实验结果表明,所提出的算法能够准确地预测深度感知质量的下降。第四章提出了一种深度视频感知优化编码算法。该算法首先通过一种低复杂度的视频质量评价方法计算虚拟绘制视点视频的空域失真和时域失真。该算法将低复杂度的视频质量评价算法计算得到的虚拟绘制视点失真作为深度视频编码率失真优化过程中的失真度量准则,并重新推导了目标函数中的拉格朗日乘子。实验结果表明所提出的深度视频感知优化编码算法可以显著地减少虚拟视点视频的时域闪烁失真。所提出的低复杂度视频评价算法对虚拟视点视频主观质量的预测准确度超过了现有的主流视频图像质量评价算法。基本上可以无缝整合到现有的三维视频编码器当中。
[Abstract]:At present, the effect of online viewing 3D video is not satisfactory for ordinary home users, not only the picture is blurred, but also the 3D stereoscopic sense is poor. This is due to the huge amount of 3D video data, the excessive compression of 3D video 3D sense and depth of experience has a great impact. On the other hand, how to evaluate the quality of 3D video is a problem that has not been solved well. The distortion type of 3D video is different from that of traditional 2D video, and the evaluation method of video quality is not good. The user quality experience of 3D video also includes more complicated factors such as depth quality of 3D video. As a kind of 3D video aided information, depth map is being used more and more widely. For 3D video system based on depth map, the quality of video image from virtual viewpoint will affect the quality of user experience of the whole system. In chapter 2, the subjective and objective quality evaluation of virtual view video with texture / depth compression distortion is studied. The design principle of the subjective data set is to ensure that the quality coverage of the virtual rendering video for testing is wide enough and that there is a certain degree of quality distinction between each other. Therefore, each texture / depth video compression quantization parameter combination is carefully selected from a large number of candidates. The virtual rendering view video subjective quality evaluation data set has been used by many famous research institutions at home and abroad. In addition, an objective quality evaluation algorithm for virtual video rendering based on full reference is proposed. The algorithm focuses on the time domain scintillation distortion caused by depth map compression and view rendering itself. The experimental results show that the proposed algorithm is superior to the existing objective video quality evaluation algorithm on the complete data set, and the superiority of the proposed algorithm is more obvious than other algorithms on the subset with obvious time-domain scintillation distortion. In the third chapter, the depth perception quality evaluation of 3D video is studied, and how image distortion affects depth perception quality of 3D video is deeply explored through subjective experiments. The experimental data set includes symmetric distortion. It also includes asymmetric distortion. The results of subjective experiments show that the loss of image details will affect depth perception. The depth perception quality scores and image quality scores obtained from subjective experiments have been published publicly. At the same time, an objective evaluation algorithm is proposed to measure the depth-sensing degradation caused by image distortion. Experimental results show that the proposed algorithm can accurately predict the decline of depth perception quality. In chapter 4, a depth video perceptual optimization coding algorithm is proposed. Firstly, a low complexity video quality evaluation method is used to calculate the spatial and temporal distortion of virtual rendering view video. In this algorithm, the virtual rendering viewpoint distortion obtained by the low complexity video quality evaluation algorithm is used as the distortion criterion in the process of the rate distortion optimization of the depth video coding, and the Lagrangian multiplier in the objective function is rededuced. Experimental results show that the proposed depth video perceptual optimization coding algorithm can significantly reduce the time domain flicker distortion of virtual view video. The prediction accuracy of the proposed low complexity video evaluation algorithm for the subjective quality of virtual view video is higher than that of the existing mainstream video image quality evaluation algorithms. Basically can seamlessly integrate into the existing three-dimensional video encoder.
【学位授予单位】：西南交通大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】