JVET-N0131 CE3: Offset-based reference sample reduction for CCLM (Tests 1.2.1 and 1.2.2) [J.-Y. Huo, X.-W. Li, J.-L. Wang, X.-Y. Chai, Y.-Z. Ma, F.-Z. Yang, B.-Z. Shen (Xidian Univ.), S. Wan (NPU), Y.-F. Yu, Y. Liu (Oppo)]

This contribution presents the results of CE3 test 1.2.1 and test 1.2.2. The two tests reduce the number of reference samples to 2 or 4 for CCLM through selecting 2 or 4 samples regularly using an offset. In test 1.2.1, the number of the reference samples is 2 or 4 according to the size of the block. In test 1.2.2, the number of the reference samples is 4, which is uniform for all block sizes.

Simulation results reportedly show BD rate on Y, Cb and Cr components and encoding/decoding time comparison over VTM4.0:

• Test 1.2.1, using 2 or 4 reference samples according to the block sizes:
  • For AI configuration: -0.03%, -0.11%, and -0.21%, with 100% EncT, 100% DecT;
  • For RA configuration: -0.01%, -0.24%, and -0.27%, with 100% EncT, 100% DecT;
• Test 1.2.2, using 4 reference samples:
  • For AI configuration: -0.03%, -0.24%, and -0.31%, with 100% EncT, 99% DecT;
  • For RA configuration: -0.01%, -0.35%, and -0.46%, with 100% EncT, 100% DecT.