Digital Signal Processing Reference
In-Depth Information
eight. For the default level, Mquant just cancels the initial
multiplication of eight.
One extreme is an Mquant of one. In this case, all AC coeffi-
cients are multiplied by 8, then divided by the quantization table
value. The results will tend to be larger, non-zero numbers, which
will preserve more frequency information at the expense of using
more bits to encode the slice. This results in higher quality
frames.
The other extreme is an Mquant of 31. In this case, all AC
coefficients are multiplied by eight, then divided by the quanti-
zation table value, then divided again by 31. The results will tend
to be small, mostly zero numbers, which will remove most spatial
frequency information and reduce bits to encode the slice
resulting in lower quality frames. Mquant provides a means to
trade quality versus compression rate, or number of bits to
represent the video frame. Mquant normally updates at the slice
boundary, and is sent to the decoder as part of the header
information.
This process is complicated by the fact that the human visual
system is sensitive to video quality, especially for scenes with little
temporal or motion activity. Preserving a reasonably consistent
quality level needs to be considered as the Mquant scale factor is
varied.
Here the state of the decoder input buffer is shown. The buffer
fills at a constant rate (positive slope) but empties discontinu-
ously as various frame size I, P, and B data is read for each frame
decoding process. The amount of data for each of these frames
will vary according to video content and the quantization scale
factors that the encode process has chosen.
To ensure the encoder process never causes the decode buffer
to over- or underflow, it is modeled using a video buffer verifier
Buffer
Fullness
Input Decoder Buffer
100%
Time
Video frame duration
( example: 1/30 Hz )
Figure 14.5. Video frame buffer behavior.
