International organisation for standardisation

5.3CE3: Motion compensation interpolation

5.3.1Summary

5.3.1.1.1.1.1.1.1JCTVC-F023 CE3: Summary report of Core Experiment on interpolation for MC [E. Alshina, T. Chujoh (CE coordinators)]

This contribution is a summary of CE3: interpolation for MC. In total 15 companies joined this CE as proponent, cross-checkers or observers. In total 21 tests were carried out and verified during this Core Experiment.

Various results were tabulated and analyzed in the summary report. Additional remarks from the discussion of the topic are recorded below.

It was acknowledged that phase shift of Motorola proposal (in JCTVC-F574) is 3/16 rather than 4/16; asserted that the effect on chroma can be fixed by modifying the chroma interpolation to align the position (proposed in JCTVC-F599, not part of CE). It was conjectured that the 3/16 shift method might work best with more reference pictures (offering phase diversity of referenceable positions).

As studied in JCTVC-F443, the 3/16 phase shift results in a greater frequency of selecting these positions relative to the use of 1/4 phase shift for these positions.

A BoG (coordinated by T. Suzuki) was asked to discuss and summarize the CE3 contributions and identify which things need further review.

It was agreed that for measuring average performance, we should weight the average according to the frequency of usage that is actually experienced when using the filter. The worst case will also be studied.