Long-Term Global Motion Compensation Applying Super-Resolution Mosaics

摘要: A new method for global motion compensation is presented significantly improving existing approaches. For each frame to be predicted a super-resolution mosaic generated that has double spatial resolution in both dimensions. The very accurate and robust mosaicing algorithm uses N preceding frames, which are available at the decoder too hybrid video coding scenario. Experimental results uncoded coded test sequences, compare our approach standard compensation. We report improvements of prediction performance terms PSNR up 7.4 dB single frames 2.22 on average. some we get decrease suggesting an adaptive algorithm.