Organisation internationale de normalisation

11.3Discussion and Conclusions

11.3.1Subset 1: Deblocking and debanding

For Subset 1, the initial viewing results seem inconsistent overall, and further viewing was deemed needed.

A revision of the summary report was made to include a report of the viewing. It was reported that the visual quality results seem very similar overall, to the degree that attempting to perform scoring appeared unnecessary because the degree of differences was so small.

As a result, it was suggested that PSNR and complexity be used for further evaluation of Subtest 1. We appear unable to confirm claims of visual quality improvement.

It was remarked that for intra coding, JCTVC-D163 has higher complexity than the others, although the proponent said that the total decoder runtime increase was only 1%. The deblocking part of the processing was reportedly increased about 20%.

None of these three proposals were suggested to be decreasing complexity.

It was asserted that there may some differences in quality that are observable from still-picture snapshots.

In JCTVC-D085, the PSNR results indicated a range from negligible to 1.6% in PSNR impact, with the MediaTek JCTVC-D163 proposal providing larger gain. The SKT/SKKU proposal affected only intra blocks, and provided about 0.5% gain on intra (with approximately no impact on other cases, as would be expected). The Microsoft and NEC proposals averaged between approximately no impact and a small degradation of PSNR. The NEC proposal adds pseudo-noise, which would not be expected to provide PSNR benefit.

There are related non-CE contributions JCTVC-D214, JCTVC-D263, and JCTVC-D377.

JCTVC-D377 reported about the same complexity as the current deblocking design, with a reported improvement of about 1.2% in PSNR performance (cross-verified and supported by the cross-verifier). The contributor of JCTVC-D377 indicated that some of the runtimes reported in their contribution were from (unreliable) cluster simulations. The cross-checker used a cluster, but a more homogeneous cluster and reported basically no change in decoding times. Several participants indicated that the changes in JCTVC-D377 were straightforward and supported its adoption at this time.

JCTVC-D214 proposed an asserted complexity reduction and parallelization improvement (making decisions based on the unfiltered signal). It was suggested that the techniques in JCTVC-D214 should hypothetically be possible to apply to a deblocking filter that has been modified as suggested in the other contributions.

JCTVC-D263 has a similar spirit to JCTVC-D214.

A suggested path was as follows: To adopt JCTVC-D377 with the parallelism improvement from JCTVC-D214 and/or JCTVC-D263, and conduct further study in a CE.

There had not been "blind" viewing of JCTVC-D377. BoG activity (Vittorio Baroncini) was requested to try to get some viewing comments and (Ken McCann) decoding times.

The proponent of JCTVC-D334 / JCTVC-C130 said that the JCTVC-D334 scheme was more similar to the current design than JCTVC-D377, although the proponent of JCTVC-D377 disputed this.

A BoG report was prepared and provided, reporting the following.

After some further blind viewing of JCTVC-D377, JCTVC-D214 and JCTVC-D263, it was reported (by V. Baroncini) that the visual difference between these and the reference was essentially imperceptible (as had been the case with the others previously), to such an extent that there appeared to be no potential benefit to trying a more extensive subjective comparison.

The decoding runtime results that were reported indicated that

• 
  JCTVC-D163 increased decoder runtime by about 1%.
  
• 
  JCTVC-D214 (parallelism) increased decoder runtime by about 6% (which may be an artifact of how it was implemented).
  
• 
  JCTVC-D263 (parallelism) increased decoder runtime by about 0.7%.
  
• 
  JCTVC-D334 increased decoder runtime by about 0.6%.
  
• 
  JCTVC-D377 had approximately no effect on decoder runtime.
  

Some participants indicated that they had wanted to study the software used in the JCTVC-D377 proposal but were not able to get access to it, as would have been enabled if a CE had been conducted for that proposal.

It was agreed that JCTVC-D377 should be further studied in a CE.

JCTVC-D214 indicates to make all horizontal and vertical decisions for the DF prior to performing the filtering (and store these decisions for use later), rather than cascading these processes.

JCTVC-D214 should be further studied in a CE.

JCTVC-D263 has two aspects – one is to shift the position of the samples that are used for decision-making so that they are not affected by the first filtering stage, and to change the filtering process to a frame-based rather than CU-based.

JCTVC-D263 should be further studied in a CE.