Of itu-t sg16 wp3 and iso/iec jtc1/SC29/WG11

6.1.7Other

JCTVC-M0111 AHG5: On chroma QP for HEVC RExt [E. François, C. Gisquet, G. Laroche, P. Onno (Canon)]

(Discussion chaired by D. Flynn.)

This contribution proposes a modification related to the chroma QP in the Range Extension of HEVC (HEVC RExt). In the current HEVC RExt Draft, one chroma QP table, linking the chroma QP (QPC) to the luma QP (QPY), is specified for each one of the three chroma formats 4:2:0, 4:2:2 and 4:4:4. In this contribution, two changes are proposed. The first change consists in removing one of the three tables, namely the 4:2:2 table. This change reportedly simplifies the design, since only the 4:2:0 table, already present in the HEVC V1 specification, and the 4:4:4 table, which is actually based on a straightforward QPC derivation, are kept. The second change consists in adding one syntax element, suggested to preferably be specified in the SPS, that indicates which table must be used in case of 4:2:2 or 4:4:4 content.

The proposal advocates the removal of the 4:2:2 chroma QP mapping table via one of two methods:

• 
  4:2:2 chroma formats use the 4:4:4 table.
  
• 
  4:2:2 and 4:4:4 chroma formats can select, via an SPS level parameter, use of either the 4:4:4 or 4:2:0 tables.
  

A custom BD-rate metric was used for some of the results that used per-component bitrates. Information on the exact allocation method was verbally reported. It was requested that this information be provided as input to the meeting.

The coding efficiency impact (when using the custom BD-rate metric) of removing the 4:2:2 table is reportedly small, mainly resulting in a small change to the balance between luma and chroma quality. However, the testing mainly focuses on the normal QP ranges, of which only two QPs (32 and 37) will be affected by the offset table.

The use of the SPS to signal the selection in the second method could be replaced by PPS signalling, however, it is not expected that this would be adapted within a sequence.

No subjecting testing has been performed, although there was some question as to there being any subjective difference. It was remarked that when the 4:2:0 tables were modified in the past, there has not been any subjective evaluation and it was questioned as to there being a need to test this one. Some experts commented that while they are open to removing the table, 4:2:2 is still an important market segment and requests further time to evaluate any impact.

It was enquired as to from where the 4:2:2 table originated. It is believed to have been a part of the initial range extension software proposal and although there is no documentation of it, that it was reported to be a middle ground between the 4:4:4 table and the 4:2:2 table.

The consensus seems to be in favour of removing the 4:2:2 table. As to whether that should provide fixed behaviour for 4:2:2, or to allow a flexible selection for both 4:2:2 and 4:4:4 is an open question. Some experts requested time to study the proposal to check that removal of the table for 4:2:2 is not going to produce any adverse effect.

Planned for further study during the next meeting cycle toward action at that time.