Page 208 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 8.5. Simulation Results 185

QSIF Foreman @ 25 f.p.s., Reflection=1.0, Contraction=0.5 QSIF Foreman @ 25 f.p.s., Reflection=1.0, Contraction=0.5
32.045 1074
1073
32.04 1072
PSNR Y (dB) 32.035 Searched locations/frame 1071

32.03 1070
1069
32.025
1068
32.02 1067
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Expansion coefficient, γ Expansion coefficient, γ
(a) Prediction quality (b) Computational complexity
Figure 8.5: Performance of SMS with di+erent values of the expansion coeDcient

(8:5), and (8:6) in Figure 8.2, because multiplications and divisions in this
case can be performed using shift operations.

8.5.1.2 Initialization, Termination, and Re1nement Tests
In order to justify di+erent parts of the SMS algorithm, the following tests
were performed.
1. Initialization Test: Two initialization procedures were tested:
(a) Random Initialization: Two of the vertices of the initial simplex
are generated randomly within the search window, whereas the third
vertex is always set to (0; 0).
(b) Proposed Initialization: This is the initialization procedure
described in Section 8.4.1.
2. Termination Test: Two termination criteria were tested.
(a) Threshold Termination: The search is terminated when the current
minimum BDMvalue is below a threshold. The threshold was set to
768, which corresponds to an average SAD=pel of 3 (16 × 16 × 3).
As already discussed, a :xed threshold is not suitable for the
BMME problem. Such a threshold does not guarantee convergence
because the global minimum BDMvalue may in some cases be
above the threshold. The threshold condition must therefore be sup-
ported by another condition to guarantee termination. In this test,
the search is also terminated if the number of iterations exceeds 10.

(b) Proposed Termination: This is the termination criterion described
in Section 8.4.3.