Graphics Reference
In-Depth Information
Boundaries where
deblocking applies
8x8 block where deblocking
can be performed
independently
Luma samples
Fig. 7.20
Illustration of picture samples, horizontal and vertical block boundaries on the 8
8
grid, and those non-overlapping blocks of 8
8 samples (marked with dotted lines ), which can be
deblocked in parallel. The dashed lines mark samples used in deblocking decisions (vertical and
horizontal)
block boundaries, the order of deblocking in each of these 8 8 deblocking units
is the same: the vertical block boundary is filtered first, which is followed by the
horizontal block boundary.
Since the HEVC deblocking can be easily parallelized, it can be done on a slice
or tile [ 16 ] basis. In this case, an encoder or decoder can choose the option to first
apply deblocking to the inner areas of a tile or slice, while leaving the deblocking on
the tile or slice boundaries. When the decoding and deblocking of all tiles or slices
is finished, the tile or slice boundaries can be processed as the last step.
Since the deblocking in HEVC is less computationally expensive and more
parallelizable than the H.264/AVC deblocking, it can be said that the deblocking
in HEVC has a better trade-off between the computational complexity, throughput,
subjective and objective quality improvements than the H.264/AVC deblocking and
is less of a bottleneck when implementing a decoder.
7.4.2
SAO Implementation Aspects and Parameters Estimation
Since SAO requires sample level operations to classify each sample into bands
or categories in both encoder and decoder, the number of operations for each
sample needs be reduced as much as possible to reduce the overall computational
complexity. At encoder-side, there are many SAO types to be tested to achieve a bet-
ter rate-distortion performance at reasonable computational complexity. Therefore,
some efficient encoder algorithms are discussed in the following sections.
 
Search WWH ::




Custom Search