Spatial scalable video coding is intended for use in applications with the compatibility, and also it has the advantages of channel adaptation and error concealment. The coding performance of the MPEG-2 spatial scalable coding is large dependent on three parts, i.e., the up conversion from the coded picture at the lower layer ("spatial prediction"), the temporal prediction from the previously coded pictures at the higher layer, and the bandwidth allocation to two layers. This dissertation presents some improvement methods for three parts in the MPEG-2 spatial scalable coding.
First, a motion-compensated up conversion method is proposed for the spatial prediction. The proposed method obtains the motion information for up conversion from the motion vectors included in the lower layer, so it has advantage in obtaining motion vectors for up conversion without extra transmission. Simulation results show that the spatial prediction by the proposed up conversion method gives a better performance than the conventional one, also that the coding efficiency of the higher layer is improved.
Second, two motion estimation methods of the higher layer are proposed in order to reduce the computational complexity for the temporal prediction. In the first one, the motion vector for the higher layer is estimated within the small search range by using the motion vector estimated in the lower layer as an initial vector. In the second one, the motion vector is estimated by spatio-temporally weighted search, based on the estimated motion vector of the lower layer. Simulation results show that the proposed methods give the smaller computational complexity without the degradation of the coding efficiency than the conventional one.
Finally, the models of distortion-bitrate relationship for the single and multiple spatial scalable sources are investigated in a fixed total bitrate system. Then the bitrate allocation methods based on the distortion-bitrate models are proposed to satisfy the user requirements of the minimum total distortion and the uniform picture quality on the constraints of bitrate. In the case of fixed bitrate for each layer, the bitrate for each picture-type is allocated to keep the constant picture quality among the picture-types. In the case of fixed total bitrate of two layers, the bitrate for each layer is allocated to give the minimum total distortion of two layers or to keep a constant distortion ratio between two layers. In the case of fixed total bitrate of multiple sources, the bitrate for each source is allocated to the minimum total distortion of the sources or to keep the uniform distortion among the sources. Simulation results show that the proposed bitrate allocation methods can well satisfy the user requirements in comparison to a fixed bitrate allocation method.
