分布式视频编码的关键技术研究

【作者】 徐祗慧；

【导师】 韩玉兵；

【作者基本信息】 南京理工大学 ， 通信与信息系统， 2010， 硕士

【摘要】 传统视频压缩技术(如H.26x等)主要在编码端进行复杂的变换、预测、熵编码等以挖掘视频的冗余信息,达到信息压缩的目的,这导致编码过程复杂,对编码器要求较高。然而,随着低成本的视频传感器网络、移动多媒体等的出现,使得传统视频编码技术已经不能满足新情况的要求。所以,分布式视频编码(Distributed Video Coding:DVC)方案开始受到了人们的关注。该编码方案将复杂的运算放在了译码端进行,从而减轻编码端的计算复杂度。本文在基于Turbo码的分布式视频编码框架的基础上,主要做了以下工作：1.阐述了Turbo码的编译码原理。重点研究了其最大后验概率译码算法,使其能更好的应用于分布式视频编译码系统的译码器中。分布式视频编码器采用Turbo码编码器,并采用基于删余矩阵的校验子请求机制,在译码失败时向译码端传送更多的校验信息。仿真结果表明,该方案的编码压缩效率高于传统帧内预测压缩编码方案(与H.263+相比较),但与传统帧间预测编码方案(双向预测)相比仍存在一定差距。2.文章研究了边信息常用的生成算法：外推插值法和内插插值法。基于本文分布式编码框架结构,我们采用内插插值法来生成边信息。文章重点分析了边信息的平均内插生成算法和基于块匹配的运动补偿内插生成算法各自的优缺点,并采用了组合算法。仿真结果表明,该方案比单纯的采用平均内插算法或单纯的采用基于块匹配的运动补偿算法取得到更好的率失真效果。3.本文还研究了分布式视频编码的虚拟信道模型,提出新的参数估计方案,仿真结果表明,在高量化比特的条件下,该方案能取得比整视频序列使用单一虚拟信道参数的更好的率失真效果。更多还原

【Abstract】 The traditional video compression technologies (such as H.26x, etc.)mainly implement complicated transformation, prediction and entropy coding at the encoder to achieve the purpose of information compression, which leads a complex encoding process and needs an encoder with a high complexity. However, with the development of the low-cost video sensor network and mobile multimedia, the traditional video coding can not meet the need of the new situation. So, the Distributed Video Coding (DVC) scheme is becoming more and more attractive.This coding scheme makes the complexity calculation on the decoder side to reduce the complexity of the encoder.Based on the Turbo codes DVC,the main works done in the paper are as follows:1.The Turbo encoding and decoding theory are expatiated. And we mainly study the Maximum A Posteriori decoding algorithm to make it more suitable to implement the decoder of DVC.Turbo encoder is used as the encoder of DVC with the difference of using request scheme based on the puncturing matrix when the decoding fails.The simulation results show that the proposed scheme has better performance than the traditional intra prediction coding scheme (H263+),but with some gap with the inter prediction coding scheme (bidirectional).2.Side information generation algorithm,such as extrapolation and interpolation, is studied.Under the framework of DVC,interpolation algorithm is used to generate the side information. Average interpolation and motion compensation interpolation based on block matching are studied. We analyze both the advantages and disadvantages of the two methods, and then, a combined algorithm is proposed. The simulation results show that the proposed scheme makes a better performance than anyone of the two methods.3.Virtual channel model in DVC is also studied in the paper, we proposed a new parameter estimation scheme, experimental result shows that the proposed scheme makes a better performance than that of using one parameter in the whole video sequence in the condition of high bit quantization.更多还原