table of contents   --> next chapter   <-- previous chapter  last modified on August 23, 2002

8. Glossary

8.1. CBR, VBR and ABR

These three acronyms stand for Constant Bit Rate, Variable Bit Rate and Average Bit Rate.

With CBR the encoder spends the same amount of bits on each frame, regardless of how much it would really need. This has the disadvantage that for example complete black pictures or complete silence in an audio frame would waste a lot of bits while pictures or audio frames with a lot of elements would benefit from these bits.

VBR now takes as much bits as it needs. So black pictures get very few bits and complex pictures get as much as they need. While this provides the best quality the final size can't be predicted at all.

This is were ABR kicks in. You specify an average bit rate you want to achieve and the encoder distributes the bits it has available. Those bits it saves on black pictures it spends on complex pictures. It is a compromise between quality and size predictability.

8.2. Global Motion Compensation (GMC)

Global motion compensation (GMC) helps in zooming (bringing the camera closer) and panning (swiveling the camera), if the objects remain static in type and only change in size and position within an image. Activation of GMC is advisable if nature films or landscape documentation are to be compressed.

8.3. Letterboxing

Perhaps you've noticed that video is sometimes surrounded by black bars prior to encoding. This process is called letterboxing. It can help to maintain the correct aspect ratios for some players. Others use the black bars for displaying subtitles. The disadvantage is that these bars enlarge the picture and therefore require more bandwidth to encode although completely black parts shouldn't need much space - but the transition between the picture and the black bars does need a lot of bits that could be used elsewhere.

8.4. P/I/B-Frames, GOPs

This was taken from
doom9's Forum:

This is a quick explanation of an MPEG "GOP", or, "Group Of Pictures".

GOP:

I frame: P frame: B frame: For every macroblock in a P frame the encoder decides whether it already knows this block from the preceding frame or whether it's completely new. In the former case it only encodes the differences (INTER mode). In the latter case it encodes the whole macroblock (INTRA mode).

The situation for B frames is like this: "I don't know that block" (INTRA mode), "I know that block from the preceding I or P frame (backward mode)", "this block looks like the block in the following frame (forward mode)" or "this block looks like a mixture of the blocks from the preceding and the following frames (bidirectional mode)".

MPEG1 GOPs are always "IBBPBBPBBPBB" (this is the order in which the frames are displayed, not necessarily the order in which they were encoded/saved). MPEG2 GOPs look like that as well. They may as well contain three B frames. The DivX5 MPEG4 codec only uses "IBPBPBPBPB" because it's easier to implement and because B frame support for MPEG4 codecs is rather new. You may expect to see more B frames in the future.

8.5. Quarter Pixel (QP)

Quarter pel or quarter pixel affects the precision in filtering of macro blocks. DivX 4 works with half pel (1.5, 1.5); 1.25, 1.75, etc. are possible starting with DivX 5. Conventional division of an image into macro blocks is refined on the basis of integers - 16x16 or 8x8 - using the supplementary information from what are known as virtual blocks. This allows the motions of objects in images to be reproduced more realistically.


table of contents   --> next chapter   <-- previous chapter  this guide was written by Moritz Bunkus