10.5 Motion estimation and compensation

The texture of each VOP is motion compensated prior to coding. The motion estimation and compensation is similar to that of H.263 with the exception that the blocks on the VOP borders have to be modified to cater for the arbitrary shapes of the VOPs. These modified macroblocks are referred to as polygons, and the motion estimation is called polygon-based matching. Furthermore, since shapes change from time to time, some conversion is necessary to ensure the consistency of the motion compensation.

A macroblock that lies on the VOP boundary, called a boundary macroblock, is padded by replicating the boundary samples of the VOP towards the exterior. This process is carried out by repetitive padding in the horizontal and vertical directions. In case there are macroblocks completely outside the VOP, they are padded by extended padding.

In horizontal padding, each sample at the boundary of a VOP is replicated horizontally in the left or right direction in order to fill the transparent region outside the VOP of a boundary macroblock. If there are two boundary sample values for filling a sample outside a VOP, the two boundary samples are averaged. A similar method is used for vertical padding of the boundary macroblocks in the vertical direction.

Exterior macroblocks immediately next to boundary macroblocks are filled by replicating the samples at the border of the boundary macroblocks. The boundary macroblocks are numbered in a prioritised order according to Figure 10.20.

Figure 10.20: Priority of boundary macroblocks surrounding an exterior macroblock

The exterior macroblock is then padded by replicating upwards, downwards, lefwards or rightwards the rows of sampling from the horizontal, vertial border of the boundary macroblock having the largest priority number. Note that the boundary macroblocks have already been padded by horizontal and vertical repetitive padding. The remaining macroblocks (not located next to any boundary macroblock) are filled with 128. The original alpha plane for the VOP is used to exclude the pixels of the macroblocks that are outside the VOP.

The reference VOP is padded based on its own shape information. For example, when the reference VOP is smaller than the current VOP, the reference is not padded up to the size of the current VOP.

The motion estimation and compensation with the padded VOPs can be carried out in several different forms, such as integer pixel motion estimation, half and quarter sample search, unrestricted motion estimation/compensation, overlapped motion compensation and advanced mode prediction. Motion vectors are then differentially encoded, similar to H.263.