Page 176 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 6.4. Discussion 153

QSIF Foreman @ 8.33 f.p.s., QP = 5, 10, 15, 20, 25, 30
38

36

34

PSNR Y (dB) 32

30

28

26 SR
SR-RC
MR
MR-RC
24
0 25 50 75 100 125 150
Bit rate (kbits/s)
Figure 6.7: R-D performance of di&erent single- and multiple-reference (with M = 50) encoders
when encoding QSIF FOREMAN at 8.33 frames=s

Due to the additional bit rate generated by the temporal components d t ,
the use of rate-constrained motion estimation and mode decision is essential
in the case of multiple-reference encoders. A single-reference rate-constrained
encoder (SR-RC) can outperform a multiple-reference encoder with no rate
constraints (MR). This is evident at very low bit rates in Figures 6.6 and 6.7
and at all bit rates in Figure 6.8. In fact, at very low bit rates, even a single-
reference encoder with no rate constraints (SR) can sometimes outperform the
multiple-reference encoder (MR).
The best overall performance is achieved by the multiple-reference rate-
constrained encoder (MR-RC). The bene)ts of this encoder become more ev-
ident as the bit rate increases. Note, however, that this improved performance
is at the expense of a signi)cant increase in computational complexity. This
increase is due to the use of more than one reference frame during motion
estimation and also to the use of rate-constrained motion estimation and mode
decision. Note, also, that at extremely low bit rates a similar performance can
be achieved by the less complex (SR-RC) encoder. Thus, at such bit rates the
use of LTM-MCP is not justi)able.