Digital Signal Processing Reference
In-Depth Information
All of these scenarios will require buffering, or temporary
storage of video files during the encoding and decoding process.
As mentioned above, this is due to the out-of-order processing of
the video frames. The larger the GOP, containing longer
sequences of I-, P- and B-frames, the more buffering is potentially
needed.
The number of bits needed to encode a given frame depends
upon the video content. A video frame with lots of constant
background (the sky, a concrete wall) will take few bits to encode,
whereas a complex static scene like a nature film will require
many more bits. Fast moving, blurring scenes with a lot of camera
panning will be somewhere in-between as the quantizing of the
high frequency DCT coefficients will tend to keep the number of
bits moderate.
I-frames take the most bits to represent, as there are no
temporal redundancy benefits. P-frames take fewer bits, and the
least bits are needed for B-frames, as B-frames can leverage
motion vectors and predictions from both previous and following
frames. P and B-frames will require far fewer bits than I-frames if
there is little temporal difference between successive frames, or
conversely, may not be able to save any bits through motion
estimation and vectors if the successive frames exhibit little
temporal correlation.
Despite this, the average rate of the compressed video stream
often needs to be held constant. The transmission channel
carrying the compressed video signal may have a fixed bit-rate,
and keeping this bit-rate low is the reason for compression in the
first place. This does not mean the bits for each frame, for
example at a 30 Hz rate, will be equal, but that the average bit rate
over a reasonable number of frames may need to be constant.
This requires a buffer, to absorb higher numbers of bits from
some frames, and provide enough bits to transmit for those
frames encoded with few bits.
14.8 Quantization Scale Factor
Since the video content cannot be predicted in advance, and
this content will require a variable amount of bits to encode the
video sequence, provision is made to dynamically force a reduc-
tion in the number of bits per frame. A scale factor is applied to
the quantization process of the AC coefficients of the DCT, which
is referred to as Mquant.
The AC coefficients are first multiplied by eight, then divided
by the value in the quantization table, and then divided again by
Mquant. Mquant can vary from 1 to 31, with a default value of
