Abstract
Different from conventional video coding, which usually employs sophisticated prediction loop to exploit the statistics at the encoder, distributed video coding (DVC) provides another way to perform efficient compression. In DVC, the statistics are exploited at the decoder to make the encoder simpler and thus it is well-suited for applications where the power and complexity are scarce sources at the encoder, e.g., wireless video phone. Wyner-Ziv video coding is a particular case of DVC that it encodes a frame independently with lossy coding, but decodes them jointly by using adjacent frame as side information. Since the rate-distortion performance is highly related to the faithfulness of the side information, the major task is how to utilize the statistics of the correlated source and generate the side information more reliably. In this thesis we propose a modified Wyner-Ziv video coding solution, using field motion-compensated interpolation method. That is, an “intra-field encoding and inter-field decoding” system. We separate each frame into one Wyner-Ziv field and one key field. Then we can use adjacent fields to interpolate the side information of each Wyner-Ziv field at the decoder. Hence a de-interlacing algorithm is employed at the decoder to exploit not only temporal but also spatial similarities between adjacent fields. By using this method, we may produce more accurate side information. Besides, the encoder can still be simple and no extra information (e.g., hash codewords) is sent from the encoder to the decoder or extra memory space at the encoder. This allows better rate-distortion performance compared to conventional Wyner-Ziv video coding scheme.