计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (1): 207-212.DOI: 10.3778/j.issn.1002-8331.1506-0244

• 图形图像处理 • 上一篇    下一篇

外推结构无反馈DVC码率控制算法研究

郑伯伟,杨春玲,欧伟枫   

  1. 华南理工大学 电子与信息学院,广州 510640
  • 出版日期:2017-01-01 发布日期:2017-01-10

Rate control algorithm for feedback-free distributed video coding based on extrapolation structure

ZHENG Bowei, YANG Chunling, OU Weifeng   

  1. School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510640, China
  • Online:2017-01-01 Published:2017-01-10

摘要: 针对现有内插结构无反馈分布式视频压缩(DVC)在大图像组条件下存在严重解码延时问题,研究了外推结构的无反馈DVC系统。在编码端,提出一种不等保护码率控制算法,该算法在编码端利用快速运动估计外推产生较高质量的边信息估计,根据比特平面误码率和比特平面重要性,实现比特平面级的码率控制;在解码端提出基于外推内插边信息更新的迭代解码优化,利用更新的边信息对每一分布式帧进行二次解码,在不增加传输码率的条件下进一步提升解码视频质量。实验结果表明,与现有无反馈码率分配算法相比,该算法能够更精确地分配码率,率失真性能提升0.6~1.8 dB,且解码视频图像的主观质量得到明显改善。

关键词: 分布式视频编码, 无反馈, 大图像组, 编码端码率控制, 边信息更新

Abstract: To avoid the severe decoding delay caused by the interpolating structure based Distributed Video Coding(DVC) under condition of long Group of Pictures(GOP), this paper studies feedback-free DVC based on extrapolation structure. At encoder, an Unequal Protection Encoder Rate Control(UP_ERC) algorithm is proposed. In which fast motion compensated extrapolation algorithm is used to obtain high quality side information estimation at the encoder, and then a bitplane-level rate control algorithm is achieved according to error probability and the importance of each bitplane. In addition, a side information refinement based on extrapolation and interpolation is proposed at decoder, utilized in the iterative decoding process to enhance the quality of the estimated frame. Experimental results show that, the proposed method provides a more precise encoding rate compared to other existing feedback-free algorithms, with a rate-distortion performance increasement by 0.6~1.8 dB. Furthermore, it also gets significant improvement on the subjective quality.

Key words: Distributed Video Coding(DVC), feedback-free, long Group of Pictures(GOP), encoder rate control, side information refinement