Multi-Hypothesis based Distributed Video Coding using LDPC Codes

Conventional video coding paradigms are usually built on the assumption that a video stream will be encoded only once but decoded several times. Over the years, this has lead to the development of encoders, which use complex motion-estimation algorithms to reduce the size of the video bitstream. However in applications like distributed sensor networks, cell-phones etc. the encoder has a limited amount of resources available to it and cannot employ complex motion-estimation algorithms. Intracoded frames used in conventional video coding have low-complexity, however they need large bandwidth to be transmitted. This has led to the development of Wyner-Ziv video codecs [1] which are based on the principles of side-information based coding, first propounded in the 1970s by Slepian-Wolf and Wyner-Ziv in their seminal work [2] and [3] respectively. Wyner-Ziv video codecs perform intra-frame encoding at the transmitter, and inter-frame decoding at the receiver. This reduces the size of the intraframe considerably, at the cost of increasing complexity at the decoder. For video transmission using cell-phones this increased decoder complexity can be absorbed by the base stations. Current distributed video coding schemes achieve compression by using a single reference frame for side-information. This single reference frame could be the temporally adjacent picture frame, or on an interpolated frame constructed using temporally adjacent frames. This paper proposes a new scheme for decoding video using more than one video frame as reference. Unlike schemes which use an interpolated frame, our decoding scheme makes use of data directly from the reference frames, giving an improvement of up to 1 dB. The approach is based on the premise that, covered and uncovered region are better represented in the original side-information frames, compared to the interpolated frame. This is especially true for sequences which do not have low-motion content. Index Terms – Channel Codes, distributed source coding, multi-hypothesis.