Databases Reference
In-Depth Information
19.10.3 Advanced Prediction Mode
In the baseline mode a single motion vector is sent for each macroblock. Recall that a mac-
roblock consists of four 8
8 luminance blocks and two chrominance blocks. In the advanced
prediction mode the encoder can send four motion vectors, one for each luminance block. The
chrominance motion vectors are obtained by adding the four luminance motion vectors and
dividing by 8. The resulting values are adjusted to the nearest half pixel position. This mode
also allows for
Overlapped Block Motion Compensation (OBMC)
. In this mode the motion
vector is obtained by taking a weighted sum of the motion vector of the current block and two
of the four vertical and horizontal neighboring blocks.
×
19.10.4 PB-Frames and Improved PB-Frames Mode
The PB frame consists of a
P
picture and a
B
picture in the same frame. The blocks for the
P
frame and the
B
frame are interleaved so that a macroblock consists of six blocks of a
P
picture followed by six blocks of a
B
picture. The motion vector for the
B
picture is derived
from the motion vector for the
P
picture by taking into account the time difference between
the
P
picture and the
B
picture. If the motion cannot be properly derived, a delta correction
is included. The improved PB-frame mode updates the PB-frame mode to include forward,
backward, and bidirectional prediction.
19.10.5 Advanced Intra Coding Mode
The coefficients for the
I
frames are obtained directly by transforming the pixels of the picture.
As a result, there can be significant correlation between some of the coefficients of neighboring
blocks. For example, the DC coefficient represents the average value of a block. It is very
likely that the average value will not change significantly between blocks. The same may
be true, albeit to a lesser degree, for the low-frequency horizontal and vertical coefficients.
The advanced intra coding mode allows the use of this correlation by using coefficients from
neighboring blocks for predicting the coefficients of the block being encoded. The prediction
errors are then quantized and coded.
When this mode is used, the quantization approach and variable-length codes have to be
adjusted to adapt to the different statistical properties of the prediction errors. Furthermore, it
might also become necessary to change the scan order. The recommendation provides alternate
scanning patterns as well as alternate variable length codes and quantization strategies.
19.10.6 Deblocking Filter Mode
This mode is used to remove blocking effects from the 8
8 block edges. This smoothing of
block boundaries allows for better prediction. This mode also permits the use of four motion
vectors per macroblock and motion vectors that point beyond the edge of the picture.
×
