Information Technology Reference
In-Depth Information
32
32
16
16
16
16
32
32
16
16
8
8
8
8
16
16
Fig. 56
Hierarchical division of motion compensated prediction blocks
16
16
32
32
Fig. 57
Diagonal block partition
Moreover, correlation between pixels in wide range can be used if transform
size is extended. Then, transform coefficients can be compacted more into the low
frequency bands. Since it is preferred to not to include block boundaries in a block
partition, it is effective to change the transform block size according to the block
partition size for motion prediction. Then, 8x8 and 16x16 block sizes are adopted
as transform block sizes and switched adaptively according to chosen motion pre-
diction block partition. The above improvements were applied to P picture, and
the computer simulation was performed using a GOP structure of IPPP. The con-
ditions of the experiment are shown in Table 7. Reference software JM15.1 of
AVC/H.264 was used as the anchor technique for the quality assessment. More-
over, the test sequences used for the experiment are shown in Fig. 58. Traffic of
Class A and Kimono1 of Class B of MPEG test sequence (YUV 4:2:0 format and
8 bpp) were chosen for the experiments. Class A includes sequences with
2560x1600 resolution that have been cropped from 4096x2048 pictures; in this
way the quality and compression efficiency of 4Kx2K could be practically evalu-
ated. Class B includes the sequences of Full HD (1920x1080) size.
The results of the experiment show the improvement in efficiency of the mo-
tion compensated prediction and the decrease of addition information such as mo-
tion vector. The amount of coding bits is successfully reduced by 2 to 30% and
PSNR was improved by 0.1-0.9 dB. The improvement of PSNR and the amount of
