Joint DC coefficient band decoding and motion estimation in Wyner-Ziv video coding

In contrast to traditional predictive coding, Wyner-Ziv video coding enables low-cost encoding architectures, in which the computationally expensive tasks for performing motion estimation are shifted to the decoder-side. In Wyner-Ziv video coding, side-information generation is a key aspect profoundly affecting the compression capacity of the system. This paper presents a novel technique which enables side-information refinement after DC coefficient band decoding in a transform-domain Wyner-Ziv video codec. The proposed side-information refinement approach performs overlapped block motion estimation and compensation, utilizing multi-hypothesis pixel-based prediction. The experimental results show that the presented Wyner-Ziv video codec incorporating the proposed technique yields significant and systematic compression gains of up to 23.22% with respect to the state-of-the-art DISCOVER codec.

1. INTRODUCTION

In order to meet the demanding prerequisites of mobility, low encoding complexity and error resilience, Wyner-Ziv (WZ) video coding - alias distributed video coding (DVC), is a recently engineered reverse paradigm anchored in distributed source coding principles (DSC) [1]. Conversely to conventional predictive video coding systems, like the H.264/AVC standard [2], which are tailored to downlink-tuned applications, DVC schemes facilitate uplink-oriented video transmission benefitting from several key advantages. In particular, DVC architectures provide error resilience and shift the computational center of gravity towards the decoder, promoting applications involving low-complexity mobile recording equipment [1]. Moreover, in a heterogeneous user environment, DVC enables adaptable systems that effectively switch encoder-decoder complexity, thereby increasing the flexibility and efficiency of video communications [3].