JVET-L0239 CE3-related: Enabling different chroma sample location types in CCLM [P. Hanhart, Y. He, Y. Ye (InterDigital)]

In VTM-2.0, the cross-component linear model (CCLM) tool assumes chroma sample location type 0 when downsampling the reconstructed chroma samples. However, other chroma sample location types should be supported. For example, in ITU-R BT.2020, it is a requirement that the chroma sample location type 2 is used. This contribution investigates the impact of the CCLM default downsampling filter on coding efficiency when encoding an input video with a chroma sample location type that is different from type 0. Simulation results for 360° content generated using chroma sample location types 1, 2, and 3 show an average coding loss in the range of 1% to 2% on the chroma components when compared to when the 360° content is generated using chroma sample location type 0. To compensate for this coding loss, this contribution proposes to use different downsampling filters in CCLM mode based on the chroma sample location type of the input video. In this way, the coding loss can be completely avoided. Simulation results for HDR/WCG content encoded using chroma sample location type 2 CCLM downsampling filter show an average coding gain in the range of -4.5% to -6.9% for the colour components based on the wPSNR metric.

Used for determining the model and for prediction itself.

An interesting problem was pointed out. Further study was recommended. E.g.

• it might be more generic to signal hor/ver offsets rather than types?
• which interpolation filters for LM determination and for prediction?

It is also noted that the number of lines for CCLM should not be increased.

Study in a CE was planned – using HDR content.