JVET-P0487 Non-CE8: Binarization of palette escape value [S. Yoo, J. Zhao, J. Nam, J. Choi, S. Kim, J. Lim (LGE)] [late]
Initial version rejected as “placeholder”.
Discussed in BoG JVET-P0999.
In VVC, the escape value is binarized with Exponential-Golomb with k equal to 3 (EG3). In the worst case, the bin length becomes 18 for a sample, which is more than fixed length coding (FL) method. Therefore, it is proposed that the escape value is binarized with truncated binary coding (TB) with a quantized maximum value. The experimental results reportedly show that the proposed scheme achieves -0.06% (AI) / 0.00% (RA) / -0.04% (LDB), respectively, when dual tree is set to 1 and under the test conditions in CE8. In the dual tree off case, the test results are -0.10% (AI) / -0.08% (RA) / 0.09% (LDB), correspondingly. For low Qp configuration, the coding performance is -1.06% (AI) / -0.47% (RA) / -0.48% (LDB) when dual tree is enabled, and -1.50% (AI) / -0.69% (RA) / -0.68% (LDB) for single tree configuration, respectively.
This contribution proposes changes to PLT escape mode binarization. It falls in the same category as in JVET-P0399. JVET-P0529 also proposes similar ideas. The three should be studied together.
The differences in this contribution are:
1) Truncated binary is used here while fixed length coding is used in JVET-P0399 and JVET-P0529.
2) No change on de-quantization in this proposal and also in one method of JVET-P0529, while a shifting based de-quantization is proposed in JVET-P0399. In JVET-P0529, an alternative method for de-quantization is proposed
It is asserted by the proponents that the main target of all 3 proposals is for simplification. Such as reduce the longest codeword length in worst case, reduce # of operations needed in de-quantization.
BoG recommendation: Study in a CE.