Video codecs, part 4: Wavelet codecs

Part 3 looks at H.264 and video over networks. Part 5 looks at Scalable video codec (SVC).

11.4 Interframe SWT Coders
The first 3-D subband coding was done by Karlson [22] in 1989. This first 3-D subband coding (Figure 11-21) used Haar or 2-tap subband/wavelet filters in the temporal direction and separable LeGall/Tabatabai (LGT) 5/3 spatial filters [25], whose 1-D component filters are as follows:

The system was subjectively optimized using scalar quantization, and DPCM to increase coding efficiency for the base subband. The 3-D frequency domain subband division of Karlson, as shown in Figure 11-21, achieves the frequency decomposition shown in Figure 11-22.

Figure 11-21. First 3-D subband filter, due to Karlson.

Figure 11-22. Frequency decomposition of Karlson's filter tree.

A simple 3-D subband coder was also designed by Glenn et al. [13]; which used perceptual coding based on the human contrast sensitivity function, but like the Karlson coder, no motion compensation. A real-time demonstration hardware of Glenn's offered 30:1 compression of SD video up to a bitrate of to 6 Mbits/sec. The visual performance was said to be comparable to MPEG 2, in that it traded slight MPEG artifacts for this coder's slight softening of moving scenes. Such softening of moving objects is typical of 3-D transform coders that do not use motion compensation.

Page 2: 11.4.1 Motion-Compensated SWT Hybrid Coding
Page 1 | 2 | 3