JVET-AC0059 AHG12: TMP using Reconstruction-Reordered for screen content coding (RR-TMP) [J.-K. Lee, D. Ruiz Coll, V. Warudkar (Ofinno)]

This contribution proposes TMP using Reconstruction-Reordered (RR-TMP) for screen content video coding. When RR-TMP is applied, the prediction signal is generated by matching the flipped L-shaped template causal neighbour of the current block with another reference block for intraTMP in a predefined search area. RR-TMP finds the best-matched flipped template position, flips the corresponding reference block according to the flip type, and uses it as a prediction block of the current block. RR-TMP can be further improved by allowing an adjusted RR-TMP block vector (BV) for intra block copy (IBC). Simulation results based on ECM-7.0 are reported below:

AI: Class F -0.19%, 0.05%, -0.17%, 100%, 101%; Class TGM -0.31%, -0.28%, -0.28%, 100%, 102%

RA: Class F -0.10%, -0.20%, -0.25%, 101%, 100%; Class TGM -0.11%, -0.22%, -0.10%, 101%, 100%

No gain was observed on camera captured content.

It was expressed by other experts that this is straightforward, gain has reasonable tradeoff with complexity and might not have much overlap with proposals adopted to next ECM. Cross-checker also supports it.

It was agreed to investigate this in an EE.