**5. Experimental results**

In this section numerous experiments with H.264/AVC reference Joint Model (JM) software version 16.1 were conducted. We compared the proposed SAAS algorithm against the FS, PIDS and UMHexagonS algorithms, in terms of computational complexity (speed measured by ME time and average search points (ASP)) and Rate-distortion performance (PSNR and bit rate). Several commonly used sequences, covering a wide range of motion characteristics, are taken into consideration.

**Figure 11.** Rate-Distortion performance comparison of FS, UMHexagonS, PIDS and SAAS at various QPs

The group of picture (GOP) structure was IPPP, in which only first frame has been coded as I frame and first P frame has been coded by UMHexagonS. The sequences are tested at 30fps (frames per second). The Content Adaptive Variable Length Coding (CAVLC) entropy coder is used for all the simulations, with 5 reference frames. A search range of 32 and the quantization parameter of 28 are used. The simulation platform in our experiments is done with a PC of 2.44 GHz CPU and 8G RAM.

190 Simulated Annealing – Single and Multiple Objective Problems

**5. Experimental results** 

are taken into consideration.

<sup>0</sup> <sup>1000</sup> <sup>2000</sup> <sup>3000</sup> <sup>4000</sup> <sup>5000</sup> <sup>6000</sup> <sup>25</sup>

Bit rate [kbp/s]

coastguard.cif

QPs

30

35

40

PSNR(dB)

45

50

FS UMHexagonS PIDS SAAS

Sequence ME TIME (sec) Average Search Points

search points reduction (%) and motion estimation time reduction (sec) (QP=28)

UMH PIDS Gain SAAS Gain UMH PIDS Gain SAAS Gain

Bus 725.6 618.3 14.79% 558.4 23.04% 30.61 19.43 37.52% 13.15 57.03% Coastguard 742.5 622.4 16.17% 569.6 23.28% 32.11 19.62 38.90% 13.29 58.62% Crew 655.5 587.1 10.43% 544.2 16.98% 20.79 14.35 30.99% 10.72 48.45% Harbour 711.8 580.8 18.40% 554.8 22.04% 33.23 18.12 45.46% 13.09 60.61% Mobile 648.7 538.7 16.97% 501.5 22.70% 28.85 16.14 44.06% 10.65 63.08% Stefan 568.8 489.5 13.94% 444.9 21.79% 25.38 15.43 39.22% 9.99 60.63% Template 597.3 512.2 14.25% 487.5 18.38% 22.40 12.53 44.07% 8.99 59.85% Average 14.99% 21.17% 40.03% 58.32% **Table 3.** Results of proposed SAAS comparing to that of UMHexagonS and PIDS in terms of average

In this section numerous experiments with H.264/AVC reference Joint Model (JM) software version 16.1 were conducted. We compared the proposed SAAS algorithm against the FS, PIDS and UMHexagonS algorithms, in terms of computational complexity (speed measured by ME time and average search points (ASP)) and Rate-distortion performance (PSNR and bit rate). Several commonly used sequences, covering a wide range of motion characteristics,

**Figure 11.** Rate-Distortion performance comparison of FS, UMHexagonS, PIDS and SAAS at various

30

35

40

PSNR(dB)

45

50

FS UMHexagonS PIDS SAAS

<sup>0</sup> <sup>1000</sup> <sup>2000</sup> <sup>3000</sup> <sup>4000</sup> <sup>5000</sup> <sup>6000</sup> <sup>25</sup>

stefan.cif

Bit rate [kbp/s]

For complexity comparisons, the proposed algorithm is compared to the hybrid UMHexagonS adopted by the H.264/AVC reference software. Two different measurements are used to calculate the computational efficiency, average search points requirement and encoding time. Results are presented in Table 3. As shown in the Table 3, SAAS needs 48- 63% less search points than UMHexagonS and saves average of 21% encoding time. Since it performs more precise search pattern adjustment, SAAS requires average 45% less search points than PIDS.


**Table 4.** Results of proposed SAAS comparing to that of FS, UMHexagonS and PIDS in terms of PSNR gain (dB) and bit-rate degradation (%) (QP=28)

Considering the large reduction in the computational complexity, the quality degradation is very small. Rate-Distortion performances are presented in Table 4. In all cases, the FS outperforms the others in image quality and bit-rate. Compared to FS, average PSNR degradation of the proposed algorithm is only 0.010. In term of bit-rate, SAAS has a slightly higher degradation than PIDS and UMHexagonS. SAAS has bit-rate decreasing of 0.64% in average. Further information can be obtained in Figure 9, which compares the ratedistortion performance among FS, UMHexagonS, PIDS and SAAS against different QPs (16, 20, 24, 28, 32, 36 and 40). Figure 10 compares the simulation results versus frame number of video sequences *Coastguard*. It is clearly reveal the superiority of SAAS to UMHexagonS in computational reduction, which more than 50% of search points are saved while the PSNR and bit-rate performance are very similar. From the results above, it can be confirmed that the SAAS algorithm has the capability to dramatically reduce the computational burden with negligible degradation in the RD performance.

Simulated Annealing for Fast Motion Estimation Algorithm in H.264/AVC 193

information. The experimental results demonstrate that more than 48% of ASP and 21% of ME time can be saved, while maintaining a similar bit-rate without losing the picture

[1] Joint Video Team of ISO/IEC MPEG & ITU-T VCEG (2004) Text of ISO/IEC 14496 10

[2] Richardson I. E. G. (2003) H. 264 and MPEG-4 video compression. The Atrium,

[3] Jianfeng R, Kehtarnavaz N, and Budagavi M(2008) Computationally efficient mode selection in H.264/AVC video coding. IEEE Trans. Consumer Electronics. 54: 877-

[4] Knesebeck M, Nasiopoulos P(2009) An efficient early-termination mode decision

[5] Hosur P. I. (2003) Motion adaptive search for fast motion estimation. IEEE Trans.

[6] Ki Beom K, Young J, Min-Cheol H(2008) Variable step search fast motion estimation for

[7] Goel S, Bayoumi M. A(2006) Multi-Path Search Algorithm for Block-Based Motion

[8] Lai-Man P, Ka-Ho N, Kwok-Wai C, Ka-Man W, Uddin Y, Chi-Wang T(2009) Novel Directional Gradient Descent Searches for Fast Block Motion Estimation. IEEE Trans.

[9] Zhibo C, Zhou P, Yun H, Chen Y(2002) Fast integer pel and fractional pel motion

[10] Zhiru S, Fernando W. A. C, Silva D. V. S. X(2010) A motion estimation algorithm based on Predictive Intensive Direction Search for H.264/AVC. in (ICME) IEEE Int. Conf.

[11] Lin C. C, Lin Y, Hsieh H. J(2009) Multi-direction search algorithm for block motion

[12] Metropolis N, Rosenbluth A. W, Rosenbluth M. N, Teller A. H, Teller E, (1953)Equation of state calculations by fast computing machines. Chemical Physics J. 21:1087. [13] Kirkpatrick S, Gelatt C. D, Vecchi M. P(1983) Optimization by simulated annealing

Southern Gate, Chichester, West Sussex, England: John Wiley & Sons.

algorithm for H.264. IEEE Trans. Consumer Electronics. 55: 1501-1510.

H.264/AVC video coder IEEE Trans. Consumer Electronics.54: 1281-1286.

Estimation. IEEE Int. Conf Image Processing. pp. 2373-2376.

estimation in H.264/AVC. Image Processing IET 3: 88-99.

Circuits and Syst. Video Technol. 19:1189-1195.

estimation. JVT 6th Meeting. JVT-F017.

Multimedia and Expo. pp. 667-672.

Science. 220: 671.

quality.

**Author details** 

**7. References** 

886.

Zhiru Shi, W.A.C. Fernando and A. Kondoz

Advanced Video Coding 3rd Edition.

Consumer Electronics. 49: 1330-1340.

*I-Lab, CVSSP, University of Surrey, Guildford, United Kingdom* 

**Figure 12.** Rate-Distortion performance and complexity cost comparison versus frame

## **6. Conclusion**

This chapter presents a novel fast motion estimation algorithm, Simulated Annealing Adaptive Search algorithm. As mv field has heavy correlation, the proposed algorithm takes the advantage of MV correlation information, which is statistically calculated and plays a significant role in SAAS process. In the SA search step, the search region is adaptively divided and the divisions are searched indexed by MV correlation probabilities in descending order. Furthermore, by utilizing Boltzmann probability concept, the minima acceptation or rejection condition of each SA search is controlled by this correlation information. The experimental results demonstrate that more than 48% of ASP and 21% of ME time can be saved, while maintaining a similar bit-rate without losing the picture quality.
