Graphics Reference
In-Depth Information
Tabl e 4. 6
Average objective and subjective
bit rate reductions provided by HEVC with
respect to JPEG and JPEG 2000 as reported in
[
6
]
Measure
JPEG (%)
JPEG 2000 (%)
BR (PSNR)
61.6
20.3
BR (MOS)
43.1
31.0
Tabl e 4. 7
Average objective
bit rate reductions provided
by HEVC with respect to
JPEG, JPEG XR and JPEG
2000 as reported in [
16
]
Quality range
JPEG (%)
JPEG XR (%)
JPEG 2000 (%)
High
49.3
32.2
26.0
Medium
58.2
39.8
30.7
Low
62.5
42.6
26.1
Similar levels of improvement over legacy codecs were reported in [
16
]usingthe
JCT-VC test set and Bjøntegaard delta metrics [
1
]. This test set includes material
mostly captured with video cameras, but it also contains some computer generated
material as well as typical screen content. Table
4.7
summarizes the results of this
study which analyzed performance at three different quality ranges: high (HEVC
QPs 12-27), medium (HEVC QPs 22-37) and low (HEVC QPs 32-47). In these
tests corresponding reference implementations were used and the resulting PSNRs
for each codec were aligned by adjusting the respective quality factors.
Coding efficiency improvements over H.264/AVC intra coding has been analyzed
for example in [
9
]. Results of that study indicate that HEVC requires on average
23 % less bit rate than H.264/AVC for still picture coding. The report also points
out that the subjective differences may be larger than that as some of the new HEVC
tools are expected to contribute more to the visual image quality than to the objective
performance of the codec.
4.6.2
Decoder Complexity
Although estimating the decoding complexity accurately can be an extremely
difficult task due to various architectural aspects depending on the implementation
platform, analyzing the traditional operational cycle counts can give reasonable
estimates about relative complexities of different algorithms.
The DC prediction, the exactly horizontal angular mode 26 and the exactly
vertical angular mode 10 in the HEVC are similar to those specified in the
H.264/AVC. Although the additional boundary smoothing operation is required in
HEVC, its impact becomes negligible when the prediction block size becomes larger
due to the smaller relative area of the filtered samples.
