PPT Slide
Modifying the block-based method
For many kinds of motion, the block-based method (pixel classifier) produces better results.
Perhaps we can make the block-based method even better.
In the Pixel-Classifier, a block of pixels in the interpolated frame is assigned a motion vector from either the forward MV calculation or the backward MV calculation. The MV with the “most overlapping area” is chosen (explain on board).
What if we chose MV that resulted in lower MAD? We’ll call this method the bi-directional copier.