TABLE OF CONTENTS Variable quantization is one of the most difficult aspects of MPEG, and is an area where active research is still being carried out. It is the most important (and most effective) of the techniques for maintaining a fixed bitrate.
The encoder must choose the quantizer_scale (and if the default table is not used, a quantization table as well); therefore, this chapter is really about the encoder decision-making process. The decoder simply follows the decision sequence defined in the compressed data bitstream.
There are several aspects to variable quantization: Bitrate control is, of course, the primary rationale for varying the quantization at all. This can be done as a global process, maintaining a fixed quantizer_scale for an entire picture and varying that scale to achieve the desired rate for that picture type. However, the quantizer scale can be adjusted at the slice layer or at the macroblock layer, thereby achieving a finer granularity in the control and (hopefully) better picture quality as well.
Aspects of variable quantization that relate primarily to bitrate control will be discussed in Chapter 15. In this chapter we assume that the basic bitrate target for a given picture has been determined. Further, the primary focus here is on control of the quantizer_scale at the macroblock level.
What remains is a difficult problem: Namely, how to distribute the bits among the macroblocks within each picture so as to obtain the best picture quality. The problem is made more difficult by the following observation:...
