The H.264 Codec - A Practical Guide to Video and Audio Compression - page 244

The de-blocking filter removes this artifact at the expense of some detail in the
final image. Because it is now defined as part of the standard, the predictive coding
process at the encoder extracts some residual information that is passed on to the de-
blocking filter as hints to assist in de-blocking the image. That way the de-blocking
process is not just a blind defocusing approach but an informed noise-reduction
technique.

13.9.3 Motion Estimation Improvements

The improved quarter-pixel resolution deployed in MPEG-4 part 2 yielded some slight
improvements in quality but these were not as good as they could be. H.264 improves the
motion estimation by changing the way that sub-pixels are interpolated.
Up to this point interpolation has been a straightforward linear proportion between
two end-point values. H.264 introduces a much more accurate interpolation of the sub-
pixel value but degenerates to the simpler interpolation if the resulting output compresses
more efficiently or accurately.
Motion estimation is applied at the 4

× 4 sub-macroblock level. A motion vector

is described for the 16 × 16 macroblock as a whole and some additional motion vectors are

4 parts of it. Those component parts within the macroblock
are organized as non-rectangular shapes within the block.

described for smaller 4 ×

13.10 Issues with GOP Structures, Frames, and Slices

The H.264 codec introduces a large number of new and complex mechanisms for referring
between frames. This makes the group of pictures (GOP) closure very hard to do unless
the encoder forces an initial I-frame and closing P-frame. The encoder must also prohibit
some of the bi-directional references to target frames outside of the GOP.

13.10.1 Temporal Prediction Improvements

The I-frame, P-frame, and B-frame model used in MPEG-1, MPEG-2, and MPEG-4 part 2
is modified considerably in H.264.
The references are now made at the slice level rather than at the frame level. The
slices are referenced in as many as 16 other frames rather than within a single frame as was
the case with the earlier codecs.
As far as the decoder is concerned, this is only a trivial addition to the decoding
complexity. At the encoding end of the process, much more information must be cached
and for a longer time. In addition, the processing overhead requires motion-compensa-
tion searches to take place within several frames where it only used to have to search
one. This is a significant computational increase and is one reason why the H.264
encoder is said to be slow compared with older codecs. When viewed in the context of
the massive improvements in bit-rate efficiency, this increase in computational load is
not as costly.