Spatio-temporal Consistency-based Wyner-ziv Video Coding Techniques And Applications

Recently, network multimedia and wireless multimedia application have signifi-cant development, e.g., video communication over3G wireless network or WiFi net-work, cell-phone video recording and sharing. Different from conventional broadcast-ing-based video communication mode, these emerging applications propose new re-quirements for video coding. For realtime video communication, error-resilient videocoding and transmission techniques are critical research directions because of lackof feedback channel or the end-to-end delay is too large to realize ARQ error resilienttechniques. In addition to the error-resilient requirements, these applications also ariserequirements for low encoding complexity due to the battery limit of handheld devices.In this paper, we will address the error-resilient video coding and low complexity videocoding problems based on the Wyner-Ziv theorem by exploiting the spatio-temporalconsistency property of natural video sequences.Based on Wyner-Ziv theory, we propose a generic error resilient video coding(ERVC) scheme using embedded Wyner-Ziv (WZ) description. At the encoder side,a joint source-channel rate-distortion optimized mode selection (JSC-RDO-MS) al-gorithm with WZ-coded anchor frames is statistically studied and developed. Givena stationary first-order Markov Gaussian source, the proposed mode optimization isjustified by an analysis of the rate-distortion (RD) impact on the WZ bit-rate. JSC-RDO-MS involves in the estimation of expected rate and distortion of WZ codingwith the unavailable side information, and the WZ bit-rate of each coding mode isdetermined based on the error correction capability of the specific WZ codec.At the decoder side, an online correlation noise model is proposed with a mixtureof Laplacians whose parameters are attained to reflect the coherence of the motion fieldof successive frames and the energy of prediction residual. Each mixture component represents the statistical distribution of prediction residuals, and the mixing coeffi-cients represent the amount of errors in motion compensation. The proposed schemeachieves the so-called classification gain by exploiting the spatially non-stationarycharacteristics of the motion field and texture. Extensive experimental results showthat the proposed WZ-ERVC scheme achieves a better overall RD performance thanexisting ERVC schemes.To improve the DVC reconstruction performance, a novel context-adaptive Mar-kov Random Field (MRF) reconstruction algorithm is developed, which exploits boththe statistical correlation and the spatio-temporal consistency by modeling the corre-sponding constraints in an MRF model, and solve the inference by finding its MRF-based MAP estimate. With the proposed reconstruction algorithm, both the subjectiveand objective reconstruction performance are improved. Parameters in this model canbe estimated with available information at the decoder side, which makes it applicablein practical applications.Finally, a novel Slepian-Wolf decoding algorithm for DVC schemes is developedto exploit not only the statistical correlation between side-information and source butalso the spatio-temporal consistency constraint of natural video sequences by model-ing the likelihood ratio information for Slepian-Wolf decoding as an anisotropic MRFmodel and solving the inference by belief propagation algorithm. The proposed de-coding algorithm is realized by iteratively implementing Slepian-Wolf decoding toimpose global bit-wise constraint introduced in the Slepian-Wolf encoding processand loopy belief propagation to enforce the local geometric regularity constraint ofnatural video frames. Compared to the existing Slepian-Wolf decoding algorithms, asignificant bit-rate saving can be achieved by the proposed algorithm.