HD-VideoBench Input Sequences and Coding Options

We have selected three resolutions that are useful for performance analysis in HD video: DVD (720x576), HD-720 (1280x720) and HD-1088 (1920x1088). The original sequences are available from TU Munchen [22] and were taken with a Sony HDW-F900 digital camera at 1920x1080 pixels resolution, 25 frames per second, progressive scan, and using a 4:2:0 chroma subsampling scheme. For the 1920x1080 resolution, we have changed the resolution to 1920x1088 in order to avoid problems with the divisibility by 16 of the image width in some encoders. Table III summarizes the main characteristics of the input sequences.
Table III: Input Sequences of HD-VideoBench
Test Sequence Resolution Frames  per second No.  frames Comments
Blue_sky
720x576 25
100 Top of two trees against blue sky.
1280x720
1920x1088
Pedestrian_area 720x576 25 100 Shot of a pedestrian area. Low camera position,
1280x720
1920x1088
Riverbed 720x576 25 100 Riverbed seen through the water.
1280x720
1920x1088
Rush_hour 720x576 25 100 Rush-hour in Munich city.
1280x720
1920x1088


The rate control mechanism used by the encoders is based on one-pass constant quality (QP) variable bit rate scheme. We do not use multiple pass or constant bit rate mechanisms because HD-VideoBench is for benchmarking the video Codecs, not the rate control algorithms. The equivalence between the quantization parameter of MPEG-2/-4 and H.264 has derived empirically (see Equation 1).
\begin{displaymath}
H264\_QP = 12 + 6 \cdot \log _2 \left(MPEG\_QP\right)
\end{displaymath} (1)

The selected sequence of frames is I-P-B-B-P-B-B. Adaptive placement of B frames is disabled. The only intra frame is the first one. We are assuming a Group of Pictures (GOP) size of 250 frames (1 second). The Motion estimation algorithms used are EPZS (Enhanced Predictive Zonal Search) [23] for MPEG-2 and MPEG-4 and hexagonal search [24] for H.264.

Mauricio Alvarez 2007-04-17