Page 132 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 4.6. Block-Matching Methods 109

Foreman @ 8.33 f.p.s. Foreman @ 8.33 f.p.s.
38 7000
8x8 8x8
36 16x16 16x16
6000
34 5000
32
PSNR Y (dB) 30 Motion overhead (bits) 4000
28
3000
26 2000
24
1000
22
20 0
1 10 20 30 40 50 60 70 80 90 99 1 10 20 30 40 50 60 70 80 90 99
Frame Frame
(a) Prediction quality (b) Motion overhead
Figure 4.4: Performance of the BMA with di erent block sizes

di erent directions. In addition, a smaller block size provides a better piecewise
translational approximation to nontranslational motion. Since a smaller block
size means that there are more blocks (and consequently more motion vectors)
per frame, this improved prediction quality comes at the expense of a larger
motion overhead, as can be seen in Figure 4.4(b). Most video coding standards
use a block size of 16 × 16 as a compromise between prediction quality and
motion overhead. A number of variable-block-size motion estimation methods
have also been proposed in the literature [98, 99]. As already discussed, the
advanced prediction mode of the H.263 standard allows adaptive switching
between block sizes of 16 × 16 and 8 × 8 on an MB basis.

4.6.3 Search Range
The maximum allowed motion displacement d m , also known as the search
range, has a direct impact on both the computational complexity and the
prediction quality of the BMA. A small d m results in poor compensation for
fast-moving areas and consequently poor prediction quality. This is evident
from Figure 4.5(a), which compares the performance of two ranges, ± 5 and
± 15. A large d m , on the other hand, results in better prediction quality but
leads to an increase in the computational complexity (since there are (2d m +1) 2
possible blocks to be matched in the search window). A larger d m can also
result in longer motion vectors and consequently a slight increase in motion
overhead, as can be seen from Figure 4.5(b). In general, a maximum allowed
5

5 As will be shown later, in block-motion elds, larger displacements are, in general, less
probable. Thus, most video codecs assign longer codewords for longer motion vectors.