【作者】 朱春波；

【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， 2008， 博士

【摘要】 随着网络技术和多媒体技术的迅猛发展,基于网络的多媒体获得了十分广泛的应用。当前的媒体应用环境具有以下几个特点:网络形式的异构性、终端设备的多样性以及多媒体应用的复杂性。这些造成了目前多媒体应用面临的困难和挑战。视频自适应技术被认为是解决这些问题并实现通用媒体访问(UMA)的最有前途的关键技术之一,是目前视频处理与通信领域研究的热点,具有重要的理论意义和广泛的应用价值。本文在基于元数据(Meta-data)空间的通用视频自适应技术的框架下,研究相应的视频自适应关键技术,主要解决异构网络对媒体访问造成的困难。本文的主要工作以及创新之处在于:1.提出了基于Meta-data空间的通用视频自适应解决方案。本文着重研究如何描述原始视频或压缩视频,使其便于自适应操作。Meta-data空间作为视频信号的特征描述集合,是视频信号的一种有效表达形式。该方案进一步提出从三个层次上解决视频自适应问题,即信号层、结构层以及语义层视频自适应。2.在信号层视频自适应中,提出了基于错误控制的视频自适应方法。异构网络的一个特点是不同形式网络中存在较大差异的通信错误概率。针对异构网络的这一特点,该方法旨在根据不同的网络条件,自适应提取压缩视频的冗余信息,使得压缩视频具有灵活的错误弹性。具体的,本文结合冗余图像编码、参考帧选择以及帧间错误跟踪,提出了一种自适应错误弹性视频编码方法,提取压缩后的冗余图像作为Meta-data,在视频通信中控制错误传播扩散。3.在结构层视频自适应中,提出了基于图像修复技术的视频自适应方案。异构网络的另一特点是不同网络间带宽变化较大,为了提高视频数据的压缩效率,传统视频编码方法往往采用粗糙的量化以满足带宽瓶颈的限制,其代价是主观视觉质量的下降。针对传统视频编码不能充分挖掘人眼视觉冗余信息的局限性,本文提取原始视频信号的纹理、结构以及运动信息作为描述原始视频信号结构层特征的Meta-data,并利用纹理合成、图像修复(image inpainting)等技术去除视频数据的视觉冗余,在不损失主观视觉质量的前提下提高了传统视频编码的压缩效率,为异构网络提供灵活带宽的视频数据。综上所述,本文对视频自适应技术进行了探讨,针对异构网络的特点在视频信号的信息提取、信息组成以及相应的自适应操作方面进行了深入的研究,取得了一些有价值的研究成果。目前基于Meta-data空间的视频自适应研究框架还处于初级阶段,尚存在一些问题,值得我们进行更深入的研究。更多还原

【Abstract】 With the development of Internet and multimedia technologies, net-based multimedia has been applied in a wide range of applications. There are three main characteristics in the current multimedia application environment: heterogeneous networks, diverse terminals and diversity of user tasks. These limitations lead to difficulties and challenges in Universal Media Access (UMA). Video adaptation is one of the most promising ways to solve these problems in UMA and is the hot spot in video processing and communication research area.Under the framework of video adaptation based on Meta-data sub-space, this thesis investigates key techniques in video adaptation and mainly focuses on overcoming difficulties in UMA which is brougnt by heterogeneous networks. The main contents and nolvelties of this paper are as follows:1. This thesis proposes a general video adaptation framework based on Meta-data sub-space.In this framework, efficient description of video signals is investigated in order to facilitate adaptation operations. As a set of feature descriptions of videos, meta-data subspace servers as an efficient representation of video signals. Furthermore, this thesis proposes to address video adaptation problems on three levels: signal level, structural level and semantic level.2. On signal level, this thesis investigates video adaptation techniques based on error control.One feature of the heterogenous networks is that various propabilities of communicating error may exist. This thesis aims to provide flexible error robustness for heterogeneous networks, by adaptively extracting redundant information from compressed videos. Specifically, a non-interactive adaptive error resilient video coding scheme is proposed combining redundant picture coding, reference picture selection and temporal error tracking. In this scheme, redundant information is extracted as the Meta-data of compressed video features to efficiently prevent temporal error propagation in video coding and communication.3. On structural level, this thesis proposes a video adaptation scheme based on image inpainting techniques.Another feature of heterogeneous networks is their fluctuating bandwidths. To improve compression ratio, traditional video coding schemes tend to use coarse quatization accuracy, which results in the degradation of visual quality. Considering that traditional video coding schemes cannot fully exploit visual redundancies in video signals, this thesis employs computer vision techniques such as texture synthesis and image/video inpainting to make use of visual redundancies. Moreover, texture, structure and motion information of the original video signals are extracted as Meta-data and incorporated into conventional video coding, in order to enhance compression efficiency without loss of visual quality.In conclusion, this thesis investigates video adaptation techniques based on extracting and building-up Meta-data and developing corresponding adaptation operations. Some useful and encouraging results have been obtained. However, up to now, the proposed framework is still at the initial stage. The video adaptation framework based on Meta-data sub-space is a potential research field which is worthy of further studies.更多还原