JVET-AC0110 Non-EE2: A Fusion method of Intra Template Matching Prediction (Intra TMP) [J.-Y. Huo, H.-Q. Du, H.-L. Zhang, W.-H. Qiao, Y.-Z. Ma, F.-Z. Yang (Xidian Univ.)]
In ECM7.0, Intra Template Matching Prediction(IntraTMP) is applied to camera-captured contents. This contribution proposes a IntraTMP Fusion method that blends several reference blocks to derive the final prediction block. These reference blocks are obtained by Block Vectors(BVs) in template matching search process. The test results on top of ECM-7.0 are as follows:
AI(Y/U/V): -0.11%/-0.11%/-0.15%, EncT 94%, DecT 102%
RA(Y/U/V): -0.xx%/-0.xx%/-0.xx%, EncT xx%, DecT xx%
Encoder time is unreliable. Cross check reports around 101%.
The search is not modified.
Five candidates are always fused, and weights are derived similar to CCCM. Usage of the fusion mode is signalled.
It was agreed to investigate this in an EE together with other fusion proposals. See notes under JVET-AC0107.