International organisation for standardisation organisation internationale de normalisation

5.16.3Coefficient level coding

All documents in this category were reviewed in a BoG (as reported JCTVC-H0724).

5.16.3.1.1.1.1.1.1JCTVC-H0724 BoG report on coefficient level coding [V. Sze]

The BoG recommended to adopt the following:

• 
  Level luma context reduction (JCTVC-H0130) (from CE)
  
• 
  Simplified coeff_abs_level_minus3 coding (JCTVC-H0498 8-bit look up table with 4 parameters)
  

Further investigation in a CE was recommended on:

• 
  Context assignment for parallel coefficient level (JCTVC-H0550) (0.1% loss but presumably better parallelism)
  
• 
  Context Set Selection for Coefficient Level Coding (JCTVC-H0404)
  

It was then concluded that H0404 should not be investigated in a CE, as the benefit for doing so is not clear from the available H0724 notes.

5.16.3.1.1.1.1.1.2JCTVC-H0062 Context Simplification on coeff_abs_level_greater1_flag [W. Ding, Y. Shi, B. Yin (BJUT)]

5.16.3.1.1.1.1.1.3JCTVC-H0690 Cross-check of JCTVC-H0062 on context simplification of coeff_abs_level_greater1_flag [H. Yang (Huawei)] [late]
5.16.3.1.1.1.1.1.4JCTVC-H0404 Context Set Selection for Coefficient Level Coding [N.Nguyen, T.Ji, D.He, G.Martin-Cocher (RIM)]
5.16.3.1.1.1.1.1.5JCTVC-H0358 Non-CE11: Cross-check of proposal H0404 on coefficient level context selection [V. Sze (TI)] [late 02-09]
5.16.3.1.1.1.1.1.6JCTVC-H0497 Improving coeff_abs_level_minus3 coding performance under high bit rate [J. Lou, X. Fang, L. Wang (Motorola Mobility)]
5.16.3.1.1.1.1.1.7JCTVC-H0607 Cross-check of JCTVC-H0497: Improving coeff_abs_level_minus3 coding performance under high bit rate [X. Zhang, S. Liu (Mediatek)] [late]
5.16.3.1.1.1.1.1.8JCTVC-H0498 Simplified coeff_abs_level_minus3 coding [J. Lou, L. Wang (Motorola Mobility)]

In the current CABAC design of HEVC, syntax coeff_ab_level_minus3 is used to represent the absolute value of a quantized transform coefficient level minus 3. Adaptive truncated Rice codes with parameters 0, 1, 2, and 3 are used. This document proposed to change the truncated Rice code with 8 bit codewords, so that memory could be saved if a lookup table is used for implementation.

This used an 8-bit codeword length limit to reduce the memory of a lookup table, and reported no loss on common conditions, although some loss at low QP (0.16 for AI HE, 0.17 AI LC, ~0.5% class F).

With parameter 4, low QP also gets gain (-0.22 for AI HE, -0.21 AI LC).

Decision: Adopt H0498, 8 bit codeword, change the Rice parameter to 4 for all cases (which causes no change for common conditions, but performs better at high bit rates).

5.16.3.1.1.1.1.1.9JCTVC-H0633 Cross-verification of simplified coeff_abs_level_minus3 coding from Motorola Mobility (H0498) [L. Guo (Qualcomm)] [late 02-04]

5.16.3.1.1.1.1.1.10JCTVC-H0550 Non-CE11: context assignment for parallel coefficient level coding [W. -J. Chien, J. Chen, J. Sole, M. Karczewicz (Qualcomm)]
5.16.3.1.1.1.1.1.11JCTVC-H0659 Non-CE11: Crosscheck for Qualcomm context assignment for parallel coefficient level coding in JCTVC-H0550 [T.-D. Chuang, Y.-W. Huang (MediaTek)] [late]