Page 160 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 5.3. E+ciency of Warping-Based Methods at Very Low Bit Rates 137

algorithm. In fact, the BMA-HO algorithm outperforms the WBA by about
0.32– 0:65 dB.
Figure 5.4 compares the subjective prediction quality of the 45 th frame of
the 8.33-frames=sFOREMAN sequence when compensated using the preceding
three algorithms. This (gure shows that BMA-HO and WBA have approxi-
mately the same subjective quality and that both outperform the BMA. More
importantly, this (gure clearly shows the type of artefacts associated with each
algorithm. The BMA su>ers from the annoying blocking artefacts. Those arte-
facts are reduced by both the BMA-HO and the WBA algorithms. However,
the BMA-HO algorithm has a low-pass (ltering e>ect that smoothes sharp
edges. This is due to the averaging (weighting) process during overlapped
motion compensation. This e>ect is very clear at the edges of the helmet. The
WBA, on the other hand, can su>er from warping artefacts. This is very clear
at the top of the helmet, where part of the helmet was stretched to compensate

th
(a) Original 45 frame of FOREMAN at 8.33 f.p.s (b) Compensated using BMA (28.06 dB)

(c) Compensated using BMA-HO (29.59 dB) (d) Compensated using WBA (29.01 dB)
Figure 5.4: Comparison between BMA and WBA in terms of subjective prediction quality