Global motion compensation

Did You Know?

• DivX offers 1 warp-point GMC encoding, which limits the global transform to panning only.
• Xvid offers 3 warp-point GMC encoding and currently has no hardware support.
• 1 warp-point GMC rarely improves video quality because panning can be described using blocks.

Operation

Global motion compensation uses a single affine transform to describe how an entire scene moves. A reference frame is panned, rotated, and zoomed according to specified warp points to generate a prediction for the next frame. Because it operates on individual pixels rather than blocks, this technique can produce predictions that block-based methods cannot achieve. In a frame using this method, each macroblock may be compensated either with global motion, requiring no additional motion data, or with local motion as if global motion compensation were not active. This choice costs one extra bit per macroblock but can enhance prediction quality and thus reduce the residual error. The transforms are only added to the bitstream when actually used, so there is no fixed bitrate overhead. A predicted frame that employs global motion compensation is known as an S-frame. A predicted frame encoded without it is called a P-frame if predicted from past frames alone, or a B-frame if predicted from both past and future frames. A frame encoded as a complete image is an I-frame.

Criticism

Global motion compensation required heavy encoding work for only small quality gains, and most of its advantages could be replicated with improved motion vector prediction. It was therefore left out of the H.264/MPEG-4 AVC standard. Because decoding it also demands significant processing power, most hardware players lack support for it.