Databases Reference
In-Depth Information
19.10.7 Reference Picture Selection Mode
This mode is used to prevent error propagation by allowing the algorithm to use a picture other
than the previous picture to perform prediction. The mode permits the use of a back-channel
that the decoder uses to inform the encoder about the correct decoding of parts of the picture. If
a part of the picture is not correctly decoded, it is not used for prediction. Instead, an alternate
frame is selected as the reference frame. The information about which frame was selected as
the reference frame is transmitted to the decoder. The number of possible reference frames is
limited by the amount of frame memory available.
19.10.8 Temporal, SNR, and Spatial Scalability Mode
This is very similar to the scalability structures defined earlier for the MPEG-2 algorithm.
Temporal scalability is achieved by using separate
B
frames, as opposed to the PB frames. SNR
scalability is achieved using the kind of layered coding described earlier. Spatial scalability is
achieved using upsampling.
19.10.9 Reference Picture Resampling
Reference picture resampling allows a reference picture to be “warped” in order to permit the
generation of better prediction. It can be used to adaptively alter the resolution of pictures
during encoding.
19.10.10 Reduced-Resolution Update Mode
This mode is used for encoding highly active scenes. The macroblock in this mode is assumed
to cover an area twice the height and width of the regular macroblock. The motion vector is
assumed to correspond to this larger area. Using this motion vector a predicted macroblock
is created. The transform coefficients are decoded and then upsampled to create the expanded
texture block. The predicted and texture blocks are then added to obtain the reconstruction.
19.10.11 Alternative Inter VLC Mode
The variable-length codes for inter and intra frames are designed with different assumptions.
In the case of the inter frames it is assumed that the values of the coefficients will be small
and there can be large numbers of zero values between nonzero coefficients. This is a result
of prediction that, if successfully employed, reduces the magnitude of the differences, and
hence the coefficients, and also leads to large numbers of zero-valued coefficients. Therefore,
coefficients indexed with large runs and small coefficient values are assigned shorter codes.
In the case of the intra frames, the opposite is generally true. There is no prediction, therefore
there is a much smaller probability of runs of zero-valued coefficients. Also, large-valued
coefficients are quite possible. Therefore, coefficients indexed by small run values and larger
coefficient values are assigned shorter codes. During periods of increased temporal activity,
