【作者】 霍俊彦；

【导师】 常义林；

【作者基本信息】 西安电子科技大学 ， 通信与信息系统， 2008， 博士

【摘要】 多视点视频是由摄像机阵列从不同角度拍摄同一场景得到的一组视频信号,包含了丰富的3D场景信息,将广泛应用于任意视点视频/任意视点电视、3D视频/3D电视等多种正在兴起的多媒体业务。然而多视点视频的数据量随着摄像机的数目增加而线性增加,巨大的数据量已经成为制约其广泛应用的瓶颈。多视点视频编码(MVC)技术主要致力于如何提高多视点视频的压缩编码效率,越来越受到学术界和工业界的密切关注。目前,MVC将作为H.264/AVC的第四个增修案,由运动图像专家组和视频编码专家组共同组成的联合视频小组正致力于MVC的标准化工作。本论文主要研究提高MVC编码效率的技术,主要研究内容及成果如下:1. MVC参考预测结构对每个视点在时间方向上采用分层B帧预测结构。根据不同时间层视点间预测对编码效率的贡献系数不同,提出了一种MVC分级预测结构。该结构对于时间相关性占主导的时间层的图像将不使用视点间预测。2.在一个GOP内,对于不同的时间层的图像进行编码时,视点间参考图像对编码效率的贡献系数随着时间层的增加而减小。据此,提出了一种灵活MVC预测结构。当前编码图像的视点间参考图像对编码效率的贡献系数可由已编码图像预测得到。如果视点间参考图像对编码效率预测的贡献系数很小,当前编码图像仅使用时间参考图像。3.与单视点视频编码标准不同,MVC在使用时间参考图像去除时间冗余的基础上增加了使用视点间参考图像去除视点间冗余的可能。MVC空间DIRECT模式照搬H.264/AVC的参考图像选取方法,忽略了MVC参考图像列表的多样性。基于此,提出了一种改进的MVC空间DIRECT模式参考图像选取方法。4.多视点视频由一组摄像机同步拍摄得到。自动白平衡是摄像机内部完成颜色补偿的主要模块,如果每个摄像机都能通过自身参数调节使得所拍摄的图像颜色逼真,则整个MVC系统所有视点的图像就达到了颜色的一致性。提出了一种基于灰度颜色点提取的自动白平衡方法,具有良好的收敛性。5.实际中,各视点之间往往存在颜色差异,降低了MVC视点间预测的准确性。提出了一种基于颜色空间转换矩阵的MVC颜色补偿方法。该方法在RGB颜色空间上建立颜色补偿模型,并进一步将颜色补偿模型扩展到YCbCr颜色空间以适用现有视频编码标准。与原始视点间参考图像相比,经过补偿后的视点间参考图像与当前编码图像更为相似,从而提高了MVC的编码效率。此外,颜色补偿模型参数可以分别从编码端和解码端推导得出,不需要将其写入码流。更多还原

【Abstract】 Multiview video captured by synchronized cameras, from different viewpoints, comprises rich 3-D information of a scene and is widely used as a signal of new types of visual media such as free viewpoint video/ free viewpoint TV and 3D Video/3D-TV. However, it results in a tremendous amount of data depending on the number of cameras. Thus, efficient compression of MVV is a key enabling factor for its widely application. Multiview video coding (MVC) has recently become an active research area, focused on compression for efficient storage and transmission of multiview video data. Currently, the standardization of MVC is being developed in the Joint Video Team (JVT) of ISO/IEC-MPEG and ITU-T-VCEG which will be an extension of H.264/AVC (Amendment 4).This dissertation deals with improving the coding efficiency of MVC, and the main research and contributions are listed as follows.1.Hierarchical B pictures are used as basic structure for temporal prediction in the reference prediction structure. Based on the analysis of the contribution of inter-view prediction to the coding gain in different temporal layer, a simplified prediction structure is proposed in which the inter-view prediction would be disabled if the picture redundancy is dominated by temporal dependency.2.The contribution of inter-view prediction to the coding gain decreases with the increase of temporal layer in each GOP. Motivated by this, the contribution of inter-view reference picture of current coding picture to the coding gain can be predicted by coded pictures. A flexible prediction structure is proposed in which only the temporal reference picture is used if the contribution derived from coded pictures is small.3.Different from the monoscopic video coding standard, both inter-view reference pictures and temporal reference pictures are used in MVC to explore the interview correlation and the temporal correlation. The reference picture selection mechanism of H.264/AVC spatial DIRECT mode is used for MVC. The spatial DIRECT mode in MVC can not work efficiently due to the variety of the reference picture list. A modified reference picture selection mechanism is proposed in the dissertation.4. Multiview video is captured by an array of cameras. Automatic white balance is one of the most important functions for each camera to provide high quality image. Color consistency among the multiple views can be achieved if each camera is white balanced. A new automatic white balance method is proposed using extracting gray color points in images, which has the advantage of robust convergence.5. The existence of color mismatch between views impairs the performance of the inter-view prediction in MVC. A novel color compensation method based on the color space conversion matrix is proposed. A color compensation model is developed in RGB channels and further extended to YCbCr channels which make the proposed method compatible to current video coding standards. The modified inter-view reference picture is more similar to the coding picture than the original inter-view reference picture, therefore the coding efficiency for MVC can be improved. Moreover, the color compensation factors can be derived both in encoder and decoder, therefore no additional data need to be transmitted to the decoder.更多还原