JVET-AM0169 Non-EE2: Reducing Candidate Modes in DDCCP [S. Wan, Y. Yin, Z. Zhu (NWPU), S. Xie, X. Zeng, C. Huang (ZTE)]

In ECM 17.0, Decoder-Derived Cross-Component Prediction (DDCCP) derives the best CCP mode from a DDCCP candidate list containing LB-CCP filtering. However, in the current ECM version, LB-CCP filtering is not used for single-model CCCM (S-CCCM), and filtering increases the complexity of the DDCCP derivation process. It is proposed to remove S-CCCM with LB-CCP and MM-CCCM with LB-CCP modes from the DDCCP list, i.e., removing the LB-CCP filtering process from DDCCP.

The experimental results on the top of ECM-17.0 are summarized as follows. The DDCCP derivation process and codec time are reduced with similar coding performance.

AI(Y/U/V): 0.02%/-0.13%/-0.11%, EncT 99.7%, DecT 99.3%

Number of modes checked in DDCCP is reduced to roughly half.

Though this is not of primary importance in this exploration, it appears straightforward and simplification confirmed by cross-checker. Some interest expressed by other experts to investigate this in an EE