多视点视频编码中关键技术的研究

【作者】 陈妤姗；

【导师】 蔡灿辉；

【作者基本信息】 华侨大学 ， 信号与信息处理， 2007， 硕士

【摘要】 随着立体显示技术的发展,有关三维(3D)视觉的研究逐渐升温,3D视频信号也成为了未来多媒体通信的主要内容。多视点视频则是现阶段表征3D视频信号的重要方式,它蕴涵了景物的深度信息,在自然场景的表征上更具真实感,在3D电视、自由视点电视、具有临场感的可视会议及虚拟现实等领域也展现了广阔的应用前景。随着视点的增加,数据压缩成了该应用领域的重点研究课题。在多视点视频中,除了各个视频流内具有很强的空间和时间相关性,各视点之间也具有一定的交叉相关性,因此,如何有效地利用视点间的视差信息以去除冗余是提高多视点视频编码效率的关键。为提高多视点视频的压缩效率,本文对多视点视频编码的预测框架、运动与视差矢量的预测、基于颜色差异补偿的视差预测编码以及基于对象的立体视频编码等方面进行了分析与研究。本文首先分析了多视点视频中视差预测特性和各种相关性的相对大小,在此基础上,提出了基于H.264的多视点视频编码方案。在视差预测中引入全局视差预测编码模式,并将其集成到H.264的多模式预测编码中,提高了压缩效率;为减少编码视差矢量和运动矢量所需的比特数,提出了改进的视差矢量和运动矢量预测方法,该方法除了利用视差矢量和运动矢量的空间相关性,还利用了它们在相邻视点或相邻时刻的对应关系。在多视点视频中,由于各摄像机所处方位不同,接收到的光线强度存在差异,同时各摄像机的增益、电平等也不能保证完全一致,导致实际获得的多视点视频图像之间存在着颜色(包括亮度和色差信号)差别,从而严重影响多视点视频的压缩性能。为进一步提高多视点视频的压缩效率,本文深入研究了基于颜色差异补偿的视差预测编码。在分析了不同视点图像之间的颜色差异基础上,对其进行建模,提出并实现了两种基于颜色差异补偿的视差预测编码方法:全局线性颜色差异补偿法及全局非线性颜色差异补偿法。实验结果验证了本文的颜色差异补偿方法明显地改善了视差预测,提高了多视点视频的压缩效率。基于对象的立体视频编码压缩技术在立体视频会议系统、视频检索等应用背景下具有较高的实用性,从立体视频信号中正确分割出立体视频对象是基于对象的立体视频编码的一个前提条件。基于此,本文分析了立体视频对象分割的特点,提出了基于对象的视差估值算法,接着通过视差分割及变化检测提取视频对象,最后设计了基于对象的视频编码方案。结果显示该方案较好地提高了压缩效率。更多还原

【Abstract】 Studies on three-dimensional (3D) vision are recently becoming increasingly popular due to advances in the 3D display technologies. In soon future, 3D video signal will be important contents of multimedia communication. As the most important 3D display technology, multi-view video has been widely used in the fields of 3D TV, free-viewpoint TV, immersive videoconferencing, virtual reality etc. However, its huge volumn of data is one major obstacle for the application of multi-view video. A multi-fold increase in bandwidth over the existing single-view makes it extremely tough to transmit and store multi-view video data. This thesis mainly concerns the problems of highly efficient multi-view video coding (MVC). To achieve high compression efficiency, correlation among different views must be exploited in multi-view video coding scheme. To this end, several efficient encoding schemes have been proposed in this paper.Firstly, an H.264-based multi-view video coding scheme is introduced. It uses the advanced predictive method of H.264 to eliminate the spatial, temporal and inter-view correlation in multi-view video. According to the characteristic of multi-view video, global disparity coding method is employed. Based on an eight-parameter global disparity model, two global disparity coding schemes are proposed. To decrease the bit rate of motion vector (MV) and disparity vector (DV), an optimized MV and DV predicted method is proposed. It utilizes not only the correlation of neighboring blocks, but also that of corresponding blocks in the adjacent images.Due to the parameter scatterance of cameras, there exists serious color-fluctuation (including brightness variation) among different views. This fluctuation will degrade the compression performance. To improve the coding efficiency of MVC, the method of color-fluctuation compensation is investigated. Firstly , color-fluctuation is analyzed based on the image formation theory. Then, several color-fluctuation compensation methods are proposed according to the simplified models. They are global linear color-variation compensation, global non-linear color-variation compensation, local color-variation compensation and global-local-adaptive color-variation compensation. Experiment results have shown these methods can greatly improve the performance of disparity prediction and the coding efficiency of MVC.Finally, object-based stereo and multi-view video technologies are discussed. An efficient algorithm for disparity matching is presented. Then, disparity field and high-order statistics based stereo video object segmentation method is proposed. Finally, an object based stereo video coder is implemented. Performance of object-based encoder verifies the validity of our algorithm.更多还原