Virtual Character Definition and Animation within the MPEG-4 Standard - 3D Modeling and Animation: Synthesis and Analysis Techniques for the Human Body - page 40

Table 1. Bit-rates [kbps] for the DCT-based coding scheme. Q denotes the
global quantization value.

Sign Q=1 Q=2 Q=4 Q=8 Q=12 Q=12 Q=24 Q=31

Bitrate[kbps] 12.08 10.41 8.5 7.29 6.25 5.41 3.95 3.33

We have tested the FBA compression techniques on a data set representing real
sign language content. In this sequence, both arms and the body of the avatar are
animated. The frame-rate for this content is 20 fps. When dealing with a wide
range of target bit rates, the DCT-based method has to be used. Since the DCT-
based method uses a 16 frames temporal buffer, an animation delay occurs. For
sign language application, this delay introduces a small non-synchronisation, but
does not affect the message comprehension. In the case of applications that
require near loss-less compression and exact synchronization of the animation
with another media, the use of the frame predictive-based method is recom-
mended. In order to increase the efficiency of the arithmetic encoder, the
MPEG-4 FBA specifications standardize a set of ranges for each animation
parameter. The global quantization step is used here for scaling the value to be
encoded in the corresponding range. Each animation parameter is encoded with
the same number of bits inside this range. If the obtained scaled value is outside
of the range, a higher quantization step has to be used.
In our tests related to sign language, when using the frame predictive-based
method, a quantization value bigger than four has to be used and the obtained bit-
rate is close to 6.2 kbps. The compression results for the DCT-based method are
presented in Table 1.
The low bit-rate, less than 15 kbps, obtained by compressing the animation
parameters, while keeping visual degradation at a satisfactory level, allows
animation transmission in a low bit-rate network.

Local Deformations

The segmented nature of an FBA compliant avatar has the main disadvantage
that during the animation seams will occur at the joints between the segments.
To overcome this limitation, a special tool based on the so-called Body Defor-
mation Tables (BDTs) has been introduced in MPEG-4. The principle consists
in adding small displacements for the vertices near the joint. Thereby, during the
animation the borders of two segments remain connected. BDTs specify a list
of vertices of the 3D model, as well as their local displacements as functions of
BAPs (ISOIEC, 2001). An example of BDTs' use is described in Preda (2002).