Digital Signal Processing Reference
In-Depth Information
chroma-prediction blocks, the sizes may be smaller depending
upon the chroma subsampling in use (4:4:4 or 4:2:2 or 4:2:0).
DC
Figure 15.3. Macroblock decomposition example for motion estimation.
15.2.7 H.264 Deblocking Filter
One of the main artifacts of MPEG-2 was blocking, or being
able to detect boundaries between the macrocell boundaries. Due
to quantization in the DCT process, many of the macroblocks
might have the same DCT coefficient values. Sometimes, nearby
blocks have sufficient difference in the pixels to cross the
threshold in the quantization process. When the differing DCT
coefficients become visually apparent, the edges of the macro-
blocks start to appear. This is especially a problem at high
compression ratios where the quantization is very coarse.
H.264 requires a deblocking filter. The deblocking filter will
smooth or filter pixel values at the macroblock boundaries for
a better visual appearance. Due to the large numbers of macro-
blocks and edges involved, this can impose a high computational
load. Various techniques are used to mitigate the number of
computations.
15.2.8 H.264 Temporal Prediction
The P- and B-frames, as in MPEG-2, are predictive frames. P-
frames utilize an earlier frame, while B-frames use both an earlier
and later frame. In H.264, multiple frames (up to a maximum of
16) can be used for prediction. The frames selected are identified
in a reference picture index. This can allow better prediction, at the
expense of much higher computations for the encoder. Of course,
the use of multiple reference frames for prediction requires the
ability to weigh, or scale the contribution from the different
frames. This provides a more accurate prediction, especially in
cases with repetitive motion, or in the situation where an object is
