Wavelet-based video coding: Optimal use of motion information for the decoding of spatially scaled video sequences

In this paper we discuss the how to best handle motion vectors in spatially scalable wavelet-based video decoders. Motion vectors with full resolution are normally included in the bit-steams relative to spatially scaled version of a video sequence. When a low-resolution version of the original sequence is received, the decoder must scale the motion vectors accordingly. We will show that the motion vector scaling (truncation) is not the best solution and that better results can be obtained by interpolating the subsampled sequence to full resolution using of the wavelet synthesis low-pass filter. We illustrate the results of experiments carried out with an in-band wavelet-based fully scalable coder that performs spatial analysis, followed by temporal filtering. Emphasis is given to the computation of the Overcomplete DWT in the spatially scalable scenario.