Back to Search Document details
25th Meeting: by teleconference, January 2022 2022-01-04 18:22
AHG9/AHG13: Film grain blending process for film grain characteristics SEI message
Authors: Yong He Qualcomm Muhammed Coban Qualcomm Marta Karczewicz Qualcomm
Abstract
This contribution proposes a film grain blending process for film grain characteristics SEI message. The blending process is identical to the process specified in SMPTE RDD5 except the modifications to reduce line buffers and on-chip memory storage.
Meeting Notes
JVET-Y0053 AHG9/AHG13: Film grain blending process for film grain characteristics SEI message [Y. He, M. Coban, M. Karczewicz (Qualcomm)]

See section 6.2.

JVET-Y0053 AHG9/AHG13: Film grain blending process for film grain characteristics SEI message [Y. He, M. Coban, M. Karczewicz (Qualcomm)]

This contribution proposes a film grain blending process for film grain characteristics SEI message. The blending process is identical to the process specified in SMPTE RDD5 except the modifications to reduce line buffers and on-chip memory storage.

The contribution proposes two independent modifications:

  1. The grain block size is changed from fixed 8x8-size in RDD5 to 8x1 for picture resolutions less than or equal to 1920x1080, 16x1 for resolution greater than 1920x1080 and up to 3840x2160, and 32x1 for resolutions greater than 4K including 8K. Using Nx1 block size eliminates the need of line buffers to compute block average in the existing methods.
  2. Add note to limit the number of film grain patterns per picture to 1 to reduce on-chip memory requirement.

Purpose of second constraint is to reduce the on-chip memory from 40 to 4 kByte.

Some concern is expressed that the proposal may be constraining too much for more generic applicability, RGB, other than 4:2:0 colour sampling, etc. Evaluation based on 1080p might not be sufficient to identify it is sufficient.

The Cross-checker is also sceptical, in particular about the modification 2. It is mentioned that the line-wise processing of modification 1 is only related to blending, the patterns can still be 2D.

Modification 1 is asserted to be uncritical in terms of subjective quality.

Should an SEI message include such constraints? Likely not. However, if such a restriction is deemed applicable and useful, the TR could express that, but then clearly mention where possible limits are.

It was suggested to discuss and study this further in the context of the AHG. It is however not practical to perform extensive tests about possible implications on subjective quality. It should, for example, be discussed if the report could point out ways to simplify the synthesis with regard to hardware complexity. If experts have an opinion that such simplifications might not be appropriate for certain type of content, this should also be expressed.

References:
JVET-W0095
Cited:
JVET-Y0009 JVET-Y0013 JVET-Y0165 X JVET-Y2023
PATENTS:
20230179805 1.00 WO/2023/107790 0.95
ARCHIVE v1 :
JVET-Y0053.docx
Decisions
See section 6.2.
It was suggested to discuss and study this further in the context of the AHG. It is however not practical to perform extensive tests about possible implications on subjective quality. It should, for example, be discussed if the report could point out ways to simplify the synthesis with regard to hardware complexity. If experts have an opinion that such simplifications might not be appropriate for certain type of content, this should also be expressed.
Citation