Digital Signal Processing Reference
In-Depth Information
present in one frame and then absent in the next, suddenly
uncovering the static background scenery.
However, most B-frames will still use only two predictive
frames, as in MPEG-2.
15.2.9 H.264 Motion Adaptive Frame and Field
Coding
When using MPEG-2, a field mode is defined using 16
8
half macroblocks. In H.264, macroblock-adaptive frame-field
(MBAFF) coding, 16
16 macroblocks are used in the field
mode. Picture-adaptive frame-field coding (PAFF) uses
a combination of frames coded as progressive or interlaced
frames, depending upon local motion, similar to motion-adap-
tive deinterlacing. This can allow individual frames, or even
macroblocks, to be encoded as interlaced or progressive,
depending on whether there are significant differences between
the top and bottom fields.
15.2.10 H.264 Spatial Prediction
Within a given frame, H.264 allows for a spatial intra-predic-
tion from the edges of neighboring blocks. The macroblock can
be predicted from other nearby macroblocks either above or to
the left. Spatial prediction of 4
4 blocks is also permitted in the
same manner. Note that MPEG-2 only allowed for DC coefficient
prediction between nearby frames.
15.2.11 H.264 Entropy Coding
Two alternatives were developed to Huffman or variable
length coding. One is content-based adaptive variable-length
coding (CAVLC). This more efficient because the coding is in
fixed tables, but adaptively changes with the data composition,
or context. The compression process of predicting, applying
DCT transforms, and quantization produces many zeros.
Run-length encoding is used to compactly represent these zero
strings. After zero,
the most
likely results to be coded
are
1.
There are several tables used to perform the variable length
coding, selected based on the data, or content. The 8
/
รพ
8 blocks
are broken into four 4
4 blocks, each of which can use
a different table, depending upon the content.
Another method used in some profiles is context-based
adaptive binary coding (CABAC). CABAC compresses data even
