JVET-AF0121 AHG12: Adjusting out-of-boundary prediction samples [P. Astola, J. Lainema (Nokia)]

This contribution proposes to improve the out-of-boundary (OOB) prediction by adjusting the uni-prediction samples using a scaling parameter and an offset to better match with the bi-prediction samples. The scale and offset are derived using the uni-prediction samples and the averaged bi-prediction samples.

The impact on coding efficiency and runtimes over ECM-10.0 is reportedly {for Y, U, V, EncT, DecT, VmPeak}: RA { -0.06, 0.03%, 0.01%, 100.3% 100.4%, 100.1% }, LB { -0.07%, 0.23%, -0.07%, 99.5%, 100.2%, 100.1%}.

The proposal tries to improve OOB prediction using scaling and offset parameters, which are derived by the encoder and the decoder in the same manner, and the scaling and offset parameters are constant for a given block.

This proposed method would replace the current OOB handling in ECM-10.0.

It was agreed to investigate this in an EE.