Organisation internationale de normalisation

Supplemental test: Combination of tests A.1 and A.2

Results from a supplemental test, which was not defined in the original CE9 description, are included here for completeness and to facilitate discussion. In this test, if the number of 16x16 sharp-edge blocks exceeds that of smooth-edge blocks, IBF is disabled for all components; Otherwise, IBF is enabled for all components.

A.1 method 1 always disables IBF, resulting in gains (0.1%...1.2%) for RGB classes, typically small losses for YUV.

A.1 method 2 makes a frame-level decision (analyzing the difference between pixels and based on this classifies the picture as sharp edge type), disabling the IBF for sharp edge types. This avoids the losses for YUV, whereas the gain in RGB is practically unchanged.

A.2 enables IBF for all three components (0.1…1.5% gain), also small gain for YUV.

A.3 modifies CCP, resulting in small changes (likely due to inclusion of ACT, gains had been larger at the last meeting).

Combination A.1/A.2 test disables for all three components when above the A.2 threshold, enabling for all three components when below. This gives more or less the same result as A.2

The differences w.r.t. compression seem to be small (in average, A.2 is 0.2-0.3% lower than A.1).

From the discussion, A.1 (adding an option to disable IBF for the first component) is asserted to be the simplest solution. However, some doubt is raised whether this would have impact on the visual quality.

Prepare viewing to compare A.1 method 1 versus A.2. If no differences are found, A.1 method 1 (sequence level disabling) should be used.

Further discussion was chaired by J. Boyce 10-23 p.m.

13.0.0.1.1.1.1.1.156JCTVC-S0309 CE9: Summary of informal subjective viewing of effects from enabling or disabling intra boundary filtering for all components [R. Cohen (MERL), P. Lai (MediaTek), J. Xu (Microsoft)]

This document provides a summary of in-meeting informal subjective testing performed for a revisit of Core Experiment 9 (CE9) on IBF/CCP interdependency. Coded videos corresponding to two tests: Test A.1 which enables the intra boundary filter for all components, and Test A.2 which disables intra boundary filtering for all components. A table of observations is included in the report.

Informal expert viewing was done, including still frames.

Decision: Adopt SPS-level disabling of IBF, as described in Method A.1.

2.