Chinese Journal of Computers   Full Text
Title	Research of the Virtual Dependency Channel in Distributed Video Coding
Authors	FANG Sheng1),2) LI Zhe1) LIANG Yong-Quan1) ZHONG Yu-Zhuo2)
Address	1)(College of Information Science and Engineering, Shandong University of Science and Technology, Qingdao, Shandong 266510) 2)(Graduate School of Shenzhen, Tsinghua University, Shenzhen, Guangdong 518055)
Year	2009
Issue	No.7(1404—1412)
Abstract &   Background	Abstract The inherent non-stationary characteristic of video signal, such as occlusion phenomena, leads to a complicated motion-compensated prediction technology. In distributed video coding (DVC) because the decoder cannot have access to the current frame, modeling the temporal correlation noise becomes a difficult task. In order to reduce the performance degradation owing to the non-stationary characteristic of video signal, this paper focuses on the research of virtual dependency channel model. Based on a common DVC codec, this paper presents a generalized model of the virtual dependency channel, analysis of which shows that in transformation domain the different subbands have various degree of sensitivity to the temporal correlation noise. According to the analysis this paper proposes a novel VCMDWT model, in which the virtual dependency channel is modeled at the LL subband with property of being more stationary, at the same time the non-stationary high frequency subbans are encoded by SPIHT with intra mode. The simulation results show that the proposed model is especially adaptable to the frames with many occlusion areas, which would be unfit for the pixel domain DVC, and finally improves the rate distortion performance with a gain up to 2.6dBs. Keywords distributed video coding; virtual dependency channel; channel model; temporal correlation noise; wavelet transform Background The work of this paper is supported by National Natural Science Foundation of China (NSFC) Key Program under contract No.90718011. As more and more mobile devices join in the networks, especially wireless networks, encoding video anywhere anytime has become a trend. An emerging approach to the low complexity video encoding is distributed video coding (DVC, also called Wyner-Ziv video coding) with side information (SI). DVC suffers from a performance gap with respect to classical motion-compensated predictive coding solutions partly on count of the difficulty in modeling the temporal correlation noise without access to current frame. In another word, DVC decoding efficiency heavily depends on the capacity to model the temporal correlation noise between the source data and the SI, which can be regarded as the input and output of a virtual dependency channel. Although many methods have been proposed to model the virtual channel well, such as building different models corresponding to the segmentation of the video areas, make the model adaptive in frame, block or pixel level, it still remains a hard challenge and does need of further research in both theory and operation. During the last years the authors have been carrying out research on improving the side information quality, the design of quantizer and the practical DVC system, and integrating the Wyner-Ziv techniques with traditional video coding system. How to generate side information with better quality is our research focus. This paper stemmed from the idea of classification-based video coding principle, which had been widely used in traditional video codec. However it is still in need of further research in DVC. It is well known the high frequency subbands is especially sensitive to inaccurate motion search, which is inevitable in DVC. The proposed VCMDWT model, in nature, belongs to classification-based coding paradigm, during which the LL subband with more stationary characteristics is encoded by Wyner-Ziv encoder, at the same time the non-stationary high frequency subbands are encoded by SPIHT with intra mode, in order to reduce the temporal correlation noise resulted from the high frequency components. The simulation results show that the proposed model is especially adaptable to the frames with many occlusion areas, and finally improves the rate distortion performance. We believe the classification-based coding paradigm will also be helpful for improving rate distortion performance of DVC. In future, we will further research how to reduce the noise according to its causes, for example, the noise from the error of dense motion fields and the noise from the translation model failure.