INTRODUCTION TO CONTENT-BASED VISUAL PROCESSING - Video Object Extraction and Representation: Theory and Applications - page 16

2. Data Granularity (Block Size)

For coding efficiency, the MPEG-1/2 standards set the basic unit
of video information, an 8x8 or 16x16 block pixels ( a macroblock ).
Although some of these constraints were based upon implementation-
dependent details such as buffer-size and microprocessor datawidth,
the optimal data size granularity reflects the effect of spatial local-
ity / correlation, independent of content.

3. Motion Estimation (I /P/B frames)

For images within a sequence, the MPEG-1/2 standards classifies the
image frames into three classes by their temporal correlation: I, P,
and B frames (short for Image, Previous and Bi-directional, respec-
tively). The encoder can describe a macroblock with reference to a
macroblock in a frame that has already been decoded, instead of its
DCT coding, to reduce bit-rate. I frames have no references to other
frames and are completely intra-coded; P frames uses macroblock ref-
erences from a previous (in the time of the video sequence) frame that
have already been decoded; B frames uses both previous and next de-
coded frames. Although 3-D DCTs compressed as well as a scheme
with these motion references, MPEG-1/2 standards chose the concept
of P,B frames because of their simplicity [Bhaskaran and Konstan-
tinides, 1995]. This method of coding also justifies our motion-based
representation as an alternate representation of the video sequence.

4. Rate Control
Rate control is an encoder-side technology that allows the encoder to
adapt its bit-rate to changing network conditions for best quality of
video and introduces feedback mechanisms and content-based quality
into the working vocabulary. In bandwidth limited environments,
rate control also implies that users prefer quality of content over
reproduction in their video playback.

For our own work, the MPEG-1/2 standard conceptually defines our
intuitive representation for our video sequence, a series of frames with
temporal references that use still image compression techniques for intra-
frame coding. The MPEG-1/2 standards are focused toward compres-
sion, independent of the content. The MPEG-4 standard drastically
changes our conception of the video sequence toward content-based pro-
cessing.

MPEG-4
The MPEG-4 standard is our reference point in our work on video
object extraction. While the MPEG-1/2 standards are primarily coding