The Community for Technology Leaders
2017 IEEE International Conference on Multimedia & Expo Workshops (ICMEW) (2017)
Hong Kong, Hong Kong
July 10, 2017 to July 14, 2017
ISBN: 978-1-5386-0561-5
pp: 357-362
Zhongyuan Wang , Research Institute of Wuhan University in Shenzhen, China
Zhiqiang Hou , NERCMS, School of Computer, Wuhan University, China
Jing Xiao , Research Institute of Wuhan University in Shenzhen, China
Ruoxi Wang , Foreign Language School Affiliated to Wuhan University, China
Rong Zhu , NERCMS, School of Computer, Wuhan University, China
ABSTRACT
Real-time video performance is difficult to guarantee under mobile communication environments characterized by loss-probability-varying transmission channel, heterogeneous networks, and various mobile devices of different access capabilities. In this paper, a novel attention-based network-aware robust video coding scheme mixing adaptive intra refresh with reference picture selection is proposed to improve the visual experience for mobile video. This method unifies human visual attention, state information of wireless channel and hybrid refresh tools to realize optimum bitrate allocation so that uneven robustness for attentive image regions against the packet loss can be readily accomplished. Further, a set of tools serving for evaluating the refresh regions, refresh modes and reference decision are developed, considering a balance among the coding efficiency, error resilience and user experience. Experiments on H.264 show that the proposed scheme provides a much better subjective feeling than traditional refresh methods when compressed video is delivered over wireless channels.
INDEX TERMS
Transceivers, Decoding, Video coding, Wireless communication
CITATION

Zhongyuan Wang, Zhiqiang Hou, Jing Xiao, Ruoxi Wang and Rong Zhu, "A hybrid intra refresh and reference selection scheme for mobile video coding," 2017 IEEE International Conference on Multimedia & Expo Workshops (ICMEW), Hong Kong, Hong Kong, 2017, pp. 357-362.
doi:10.1109/ICMEW.2017.8026249
83 ms
(Ver 3.3 (11022016))