Page 213 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

190 Chapter 8. The Simplex Minimization Search

motion was estimated using macroblocks of 16 × 16 pels, a maximum allowed
displacement of ± 15 pels, SAD as the distortion measure, restricted motion
vectors, and full-pel accuracy. In this case, however, motion vectors were
predictively encoded using the median prediction method and the VLC table
of the H.263 standard. In addition, motion was estimated and compensated
using reconstructed reference frames rather than original frames. Both, the
frame signal (in case of INTRA) and the DFD signal (in case of INTER)
were transform encoded according to the H.263 standard. To generate a range
of bit rates, the quantization parameter QP was varied over the range 5–30
in steps of 5. This means that each algorithm was used to encode a given
sequence six times. Each time, QP was held constant over the whole sequence
(i.e., no rate control was used). The :rst frame was always INTRA encoded,
and all other frames were INTER encoded. No INTRA=INTER switching was
allowed at the macroblock level. The INTRA bits were included in the bit-rate
calculations, and no header bits were generated. All quoted results refer to the
luma components of sequences.
Figures 8.7 and 8.8 show examples of the rate-distortion (R-D) perfor-
mance of the SMS algorithm and compare it to that of the other :ve BMME
algorithms. Figure 8.7 shows the results for the FOREMAN sequence with frame
rates of 25 frames=s and 8:33 frames=s, whereas Figure 8.8 shows the results
for the AKIYO and TABLE TENNIS sequences with frame rates of 10 frames=s and
15 frames=s, respectively. Both :gures con:rm the superior R-D performance
of the SMS algorithm compared to other fast BMME algorithms.
The superior performance of the SMS algorithm is also shown on a frame-
by-frame basis in Figure 8.9. This :gure shows the performance for the
FOREMAN sequence at 8:33 frames=s with a quantization parameter of QP = 10.
For clarity, the :gure shows only the performance of the FS, SMS, NSS, and
OTS algorithms. As can be seen, the SMS algorithm provides the closest pre-
diction quality (Figure 8.9(a)) to the FS algorithm. This results in the use of
fewer bits for the DFD signal (Figure 8.9(c)). In addition, the initialization pro-
cedure results in less motion overhead (Figure 8.9(d)). The reduced number of
DFD bits and motion bits results in a reduced overall bit rate (Figure 8.9(e)).
This is all achieved at a reduced computational complexity (Figure 8.9(b)).

8.5.3 Results Within an MPEG-4 Codec

In a collaborative work, the SMS algorithm has also been tested within an MPEG-
4 codec. The results in this subsection are reproduced, as is, from Ref. 175. 5

5 The authors would like to thank Mr. Oliver Sohm for incorporating SMS within MPEG-4 and
providing the results.