基于视频信号描述的视频自适应技术研究

【作者】 王毅；

【导师】 刘政凯； 李厚强；

【作者基本信息】 中国科学技术大学 ， 信号与信息处理， 2007， 博士

【摘要】 基于网络的多媒体应用是当前多媒体技术发展的必然趋势。多媒体应用环境的多样性给终端用户提供了灵活便捷的多媒体信息获取方式，提高了用户的工作效率和生活质量。但是，这种多样性同时也造成了媒体信息访问的困难。视频自适应技术是目前解决这些问题实现通用媒体访问最有前途的研究方向。本文将视频自适应技术作为研究内容，对这领域进行了一些探索性的研究工作。首先，本文对现有的视频自适应研究框架进行了分析，指出该框架存在的问题——不利于通用高效视频自适应技术的设计。为克服这一缺点，本文提出了一种新的研究框架以及相应的算法，主要创新之处有：①本文提出了基于视频信号描述的自适应技术研究方案，即着重研究如何描述原始视频信号使其便于自适应操作。本文进一步提出用特征描述集合的形式作为视频信号的表达形式，每种特征描述反映了原始视频信号某些方面的特性。利用这些特性，可以设计简单高效的自适应操作。②基于上述研究框架，本文提出一种运动信息描述算法用于实现码率的自适应功能。我们首先提出了一种基于分块模式的分层模型用于描述运动信息，然后提出了一种预编码的算法获取运动信息，最后还提出了运动信息的快速抽取算法。本算法生成的运动信息描述可用于码流的快速生成。当需要某一码率的视频时，可以从运动信息描述中抽取出对应的运动信息用于编码，由于不需要运动估计，所以可以快速生成码流。③本文提出一种率失真信息的提取算法可应用于码率分配。我们首先分析了预测帧编码的依赖性提出了一种依赖性的线性描述模型，并定义了影响因子的概念。在此基础上，我们提出了单帧率失真函数的修正算法。修正后的函数隐含了依赖性的考虑，反映了整体失真和单帧码率的关系。我们将修正后的率失真函数应用于码率分配，提高了整体的编码性能。④本文提出一种基于感兴趣信息的空间自适应解决方案。移动终端用户由于有限的显示屏尺寸在欣赏高分辨率视频时存在困难。我们基于感兴趣区域的思想，提出将高分辨率视频转换成由感兴趣区域组成的低分辨率视频以克服显示屏尺寸带来的限制，同时，尽可能保留原视频中的信息，提高用户的欣赏效果。整个方案包括感兴趣信息的提取和封装算法、基于感兴趣信息的量化参数自适应调整算法和快速模式选择算法。综上所述，本文对视频自适应技术进行了深入地探讨，取得了一些有价值的研究成果。目前，本文提出的视频自适应的研究框架还处于初级阶段，存在很多问题有待解决，值得我们进行更深入地研究。更多还原

【Abstract】 Internet based multimedia application is the evolution trend of multimedia techniques. Current complex media environment provides users flexible media data access, which improves working efficiency and life quality of users. However, the diversity property also brings serious difficulties in media data access. Video adaptation is the most promising solution for achieving universal media access.We studied video adaptation techniques in this dissertation. Firstly, we analyzed existing video adaptation framework and found that it is not appropriate for designing general and efficient video adaptation operations. Based on this analysis, we proposed a novel video adaptation framework and some related algorithms. The main contributions of this dissertation can be summarized as follows.①We proposed a video adaptation framework based on video signal representation, in which the key point is how to describe video signal to facilitate it in video adaptation operations. We also proposed the set of feature descriptors as the representation of video signal. Each descriptor reflects some characteristics of original video signal. By making using of these descriptors, we can develop simple and efficient video adaptation operations.②Based on the framework above, we proposed a motion description algorithm for bit-rate adaptation. Firstly, we proposed a mode-based hierarchical model for motion information. Secondly, we proposed a pre-encoding method to achieve motion descriptor. We also developed a method of motion extraction. The generated motion descriptor can be applied in fast bit stream generation. While encoding a video in some bit-rate, the encoder can extract proper motion information to accelerate the encoding process. Due to skipping motion estimation, the encoding complexity can be reduced greatly.③We proposed an algorithm to extract rate-distortion information from original video signal. The information can be used in bit allocation. We analyzed the dependency among predictive frames and proposed a linear model for it. We also introduced the definition of impact factor. Based on the linear model, we proposed a method to modify the rate-distortion function of each frame. Then, the modified functions with dependency consideration reflect the relationships between the whole distortion and the single frame rate. With these modified functions, it is easy to achieve optimal bit allocation to improve the encoding performance.④We proposed a spatial adaptation framework based on attention information. The limited display size of mobile devices hinders the perceptive experiences of users when they browse high-resolution videos. Based on the idea of ROI, we proposed to transform high-resolution videos into low-resolution ones which are composed of attention areas in each frame, which will meet the constraint from the limited display size. At the same time, most of attention information in original video signal will be remained. With this framework, the perceptive experience of mobile users will be improved. The whole framework includes three algorithms: the extraction and encapsulation of attention information, QP adaptive adjustment based on attention information, and fast mode decision for transcoding.In conclusion, we studied video adaptation techniques and achieved some valuable results. Till now, the proposed framework is still at the initial stage and there exist many problems needed to be solved. The video adaptation framework based on video signal representation is a potential research field which is worthy of further study.更多还原