Proposed FME Mode Pre-selection Algorithm

Fig. 4.19 shows the proposed FME mode pre-selection concept for SVC. In our proposed FME mode pre-selection algorithm, the IME is executed for the prediction modes of Inter, Inter+residual, Inter-layer motion, and Inter-layer motion+residual first. Afterwards, the proposed FME mode pre-selection algorithm is applied for all prediction modes coming from the results of IME to skip the potentially ignorable prediction modes before entering the FME operation. Once the candidate prediction modes have been decided by the proposed FME mode pre-selection algorithm, the selected candidate modes will be fed into the FME module to choose the best prediction mode.

IME for Inter mode

Mode filtering for FME IME for Inter

mode + Residual IME for ILM IME for ILM + Residual

Best mode selection Best mode

16x16 16x8 8x16 Sub-mode

FME

Fig. 4.19. Illustration of mode FME mode pre-selection for SVC

Fig. 4.20 shows the flowchart of our proposed mode pre-selection algorithm which includes four types of mode pre-selection algorithm. In this flowchart, the candidate set of prediction modes Φ is defined as follows.

Φ _ij (4-22)

105

Initialization ∀Φij=True|Φij∈

Φ

Execute FME for

Φ

Type1 FME mode pre-selection for

Φ

Type2 FME mode pre-selection for

Φ

Type3 FME mode pre-selection for

Φ

Type4 FME mode pre-selection for

Φ

Fig. 4.20. Flowchart of proposed FME mode pre-selection algorithm for SVC

Fig. 4.21 shows the detailed flowchart of proposed FME mode pre-selection algorithm for Type1. In this flowchart, the weightings of ω1 and ω2 are computed first. Afterwards, the IME rate distortion cost relationship between Inter and Inter-layer motion is compared for all block size one by one to skip the potentially ignorable prediction modes. Once the Type1 mode pre-selection algorithm has been done, the candidate set of Φ will be fed into the next mode pre-selection algorithm to further pre-select possible modes. Fig. 4.22 exhibits the detailed flowchart of proposed FME mode pre-selection algorithm for Type2. The flowchart of Type2 mode pre-selection algorithm is very similar to the mode pre-selection algorithm of Type1 except that the rate distortion cost relationship is compared between Inter+residual and Inter-layer motion+residual.

106

Fig. 4.21. Detailed flowchart of proposed Type1 mode pre-selection algorithm

107

Fig. 4.22. Detailed flowchart of proposed Type2 mode pre-selection algorithm

Fig. 4.23 and Fig. 4.24 show the detailed flowchart of proposed Type3 and Type4 mode pre-selection algorithm, respectively. From these figures, the most determination processes are similar to the determination processes of Type1 and Type2 mode pre-selection algorithms.

However, there are some addition determination processes labeled by bold line have been further included. The additional included determination processes here are in charge of avoiding unnecessary determination operations since some candidate prediction modes might have been already disabled in the previous Type1 and Type2 mode pre-selection algorithms.

Therefore, by adding additional determination processes into the mode pre-selection flowchart, the unnecessary operations can be avoided and thus achieve computational complexity saving.

108

Fig. 4.23. Detailed flowchart of proposed Type3 mode pre-selection algorithm

109

Fig. 4.24. Detailed flowchart of proposed Type4 mode pre-selection algorithm

110

4.4. Simulation Results

In this section, several simulation results are shown to demonstrate the performance of our proposed FME mode pre-selection algorithm. The simulation settings are summarized in Table 4-VI.

Table 4-VI Simulation settings

Reference software JSVM9.17 [63]

QP for spatial base layer 18, 28, 33, 38 QP for Spatial enhancement layer 12, 22, 27, 32 Frame size in spatial base layer QCIF and 540P Frame size in spatial enhancement layer CIF and 1080P

Frames to be encoded 300 and 150

540P and 1080P Blue_sky, Pedestrian, Riverbed, Station2, Sunflower, Tractor

Table 4-VII shows the BD-PSNR and BD-Rate comparisons for our proposed FME mode pre-selection algorithm subject to full mode FME. In this table, the quantization parameter values of 12, 22, 27, and 32 are adopted to derive the results. It should be mentioned that since the only highest quality (including spatial, temporal, SNR) layers would be decoded from the SVC bitstream, we only show the QP values for highest spatial layers. For QCIF and CIF case, the average BD-PSNR degradation and BD-Rate increasing is 0.034 and 0.0777%, respectively. In addition, for the 540P and 1080P case, 0.037 and 0.914% BD-PSNR degradation and BD-Rate increase can be achieved. From this table, it is obvious that our proposed FME mode pre-selection algorithm results in ignorable rate distortion performance loss when compared to full mode FME. Table 4-VIII further shows the detailed PSNR degradation and bitrate increasing for different QP values. On average, our proposed algorithm only results in 0.005dB PSNR degradation and 0.89% bitrate increasing.

111

Table 4-VII BD-PSNR and BD-Rate comparisons for proposed algorithm subject to full mode FME

Resolution Sequences BD-PSNR BD-Rate (%)

BL: QCIF

Table 4-VIII Rate distortion performance comparisons for proposed algorithm PSNR degradation (dB) Bitrate increase (%)

QP=12 QP=22 QP=27 QP=32 QP=12 QP=22 QP=27 QP=32

112

Table 4-IX tabulates the mode reductions of our proposed algorithm. On average, our proposed algorithm can achieve 65.97% mode reduction whatever what quantization parameters have been used.

Table 4-IX Mode reductions for proposed algorithm subject to full mode FME

(%) QP=12 QP=22 QP=27 QP=32

Fig. 4.25 reveals the hardware architecture design of our proposed FME more pre-selection algorithm. When designing a modern video encoder system, the IME and FME are usually separated into two individual pipeline stages as Fig. 4.25(a) shown due to the reason of balancing the computational loads for each stage. Therefore, the proposed FME mode pre-selection algorithm should be placed in the middle of two pipeline stages. However, there are two places where we can use for arranging our proposed FME mode pre-selection algorithm. The first place is at the same pipeline stage of IME and the alternative place is at