Information Technology Reference
In-Depth Information
method which increases operation accuracy is used. It is supposed that gains
of about 4 to 7% of improvement are realized.
(2) Transform
The size of a macroblock is expanded to maximum of 64x64, and the trans-
form of several block sizes from 4x4 to 16x16 is available. It is effective for
high resolution video. It has been claimed that improvements in the range of
10 to 15% for P picture and 15 to 20% for B picture can be achieved.
(3) Loop Filter
The Wiener filter which performs image restoration from a local decoded
picture using source picture is designed. The filter can be also turned on and
off for each block to improve the quality of a local decoding picture. Gains
of about 5 to 10% of improvement have been reported.
(4) Intra Coding
Prediction between pixels is alternatively performed from several directions
in the Intra coding of AVC/H.264. In this tool, transform basis is changed
according to the directions of the predictions. Improvements of about 7%
have been shown.
These tools are the improvements to the existing AVC/H.264 framework and keep
the existing coding architecture. They adopt the approach of changing the range of
parameters, selecting adaptive case among several candidates and performing the
optimization which could not be employed by the restriction of practical memory
and operation scale before. It is said that KTA could provide about 20 to 30% of
performance improvement relative to AVC/H.264 by adoption of these tools.
6.3.1.2 Block Size Extension and Non-Rectangle Block Application
A coding method which extends AVC/H.264 and performs UHD video coding us-
ing block size extension and a non-rectangle block is introduced as an example
[31]. Macroblock size is extended from 16x16 to 32x32, and accordingly the block
partitions for motion prediction are expanded to 32x32, 32x16, 16x32 and 16x16.
When 16x16 block partition is chosen, sub-block partitions of 16x16, 16x8, 8x16
and 8x8 are also employed. Smaller block partitions of 8x4, 4x8 and 4x4 are not
used because noise components will become dominant and the essential structure
information of a picture will become impossible to be expressed efficiently by
such small blocks in the case of UHD video coding.
In addition to extension of macroblock size, non rectangle block partitions are
employed for motion prediction. Although non rectangle block partition was pro-
posed for the purpose of expressing more complicated motion with a short motion
vector in super low bit rate video coding, it is believed that new partitions will con-
tribute to reducing residual energy by effectively expressing complicated picture
structures such as object boundaries in UHD video. On the other hand, there are
some problems such as the increase of the motion detection operations, the increase
of the shape description information and the necessity of the memory access to
complicated shape. Then, the simple diagonal partitions shown in Fig. 57 are
adopted, which can be created with the combination of 16x16 blocks.
Search WWH ::




Custom Search