What are some myths about MPEG?
A. There are a few major myths that I am aware of: 1. Block displacements: macroblock predictions are formed out of arbitrary 16×16 (or 16×8/16×16 in MPEG-2) areas from previously reconstructed pictures. Many people believe that the prediction macroblocks have boundaries that fall on interchange boundaries (pixel 0, 15, 31, 53… line 0, 15, 31, 53… etc.). In fact, motion vectors represent relative translations with respect to the target reconstruction macroblock coordinates. The motion vectors can point to half pixel coordinates, requiring that the prediction macroblock to be formed via bi-linear interpolation of pixels. 2. Displaced frame (macroblock) difference construction: the prediction error formed as the difference between the prediction macroblock and source macroblock is coded much like an Intra macroblock. The prediction may come from different locations (as in bi-directional prediction–or in MPEG-2–16×8, field-in-frame, and Dual Prime), but the DFD is always coded as a
