INTERNATIONAL ORGANISATION FOR STANDARDISATION ORGANISATION INTERNATIONALE DE NORMALISATION ISO/IEC JTC1/SC29/WG11 CODING OF MOVING PICTURES AND AUDIO

Independent work of different research groups has recently demonstrated that the conventional "t+2D" architecture which is currently investigated in the SVC standardization may provide limited efficiency and flexibility with respect to resolution scalability. A significant part of the problem lays in the fact that, although the temporal motion-compensated transform is applied in the full-resolution input frames during encoding (i.e. prior to the spatial transform), low-resolution decoders cannot replicate this process since certain high-resolution, high-frequency content is missing. More importantly, as the motion vectors are estimated at full resolution, the shift uncertainty introduced during the motion-vector downscaling process (in order to make vectors applicable at low-resolution frames) can produce visual artifacts in cases where motion is modelled incorrectly. In this contribution, we start from the basic concepts that justify the fundamental operation of a "2D+t" architecture, which is motion compensation making use of the Overcomplete Discrete Wavelet Transform (ODWT). We then review our previously-proposed architecture for the efficient performance of motion- compensated temporal filtering (MCTF) in the wavelet domain. Experimental results indicating the problems of the "t+2D" architecture for low-resolution decoding, as well a comparative test between "t+2D" and "2D+t" systems under the same framework of advanced MCTF, substantiate the performance of the proposed architectures.