Page 134 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 4.6. Block-Matching Methods 111

Foreman @ 8.33 f.p.s.
33
1/2-pel, full search
1/2-pel, refinement
full-pel, full search
32

31

PSNR Y (dB) 30

29

28

27

26
1 2 3 4 5 6 7 8 9 10
Frame
Figure 4.6: Performance of the BMA with subpel accuracy

more possible candidate blocks within the search window. For example, when
moving from full-pel to half-pel accuracy, the number of candidate blocks in
2
the search window increases from (2d m +1) to (4d m +1) . To alleviate this
2
complexity, most video codecs implement subpel accuracy as a postprocessing
stage, where rst a full-pel motion vector is obtained, usually using full search,
and then this vector is re ned to subpel accuracy using a limited search. This
provides a large saving in computational complexity and at the same time
maintains the improved prediction quality, as can be seen in Figure 4.6.

4.6.5 Unrestricted Motion Vectors

In some cases (like, for example, in border blocks) part of the search window
is outside the reference frame area. This means that some of the candidate
blocks in the search window are either partially or completely out of the
reference frame. There are two ways to handle such candidate blocks. In the
restricted motion vectors method, such blocks are ignored and skipped during
motion estimation. In the unrestricted motion vectors method, however, such