Game Development Reference
In-Depth Information
parameter in frame n-1 is used as a prediction. Depending on precision
requirements on FAPs/BAPs, different quantization step sizes could be applied.
They consist of a local (FAP/BAP specific) step size and a global one (used for
bit-rate control). The quantized prediction error is then encoded by arithmetic
coding. Taking into account the fact that natural human motions are constrained
by physical properties, the range of each FAP/BAP is limited to a specific
interval. Using this property, the coding efficiency is increased.
The DCT-based coding method (Figure 7b) splits FAP/BAP time sequences into
segments made of 16 consecutive frames. Three steps are necessary to encode
each segment: (1) the determination of the 16 coefficient values using discrete
cosine transform (DCT); (2) quantizing and coding the alternative coefficients
(AC); and (3) predictively coding and quantizing the continuous component (DC)
coefficients. The global quantization step Q for the DC coefficients can be
controlled and the AC coefficients quantization step is set to 1/3 of Q. The DC
coefficients of an intra-coded segment are encoded as it is and, for an inter-
coded segment, the DC coefficient of the previous segment is used as a
prediction of the current DC coefficient. The prediction error and alternative
component coefficients (AC), (for both inter and intra-coded segments), are
coded using Huffman tables.
The current FBA encoder implementation from the MPEG reference soft-
ware, 12 as well as commercial 13 and academic implementations (Capin, 1999;
Preda 2002) shows a very low bit rate for compression of the animation
parameters, ranging from 2kbps for the face, up to 40 kbps for the entire body.
Figure 7. Decoding block diagram.
(b) DCT-based method.
(a) Frame predictive-based method.
Search WWH ::




Custom Search