Back to Search Document details
15th Meeting: Gothenburg, SE, July 2019 2019-06-25 19:44
Non-CE5: Simplification on clipping value calculation for adaptive loop filter
Authors: Nan Hu Qualcomm Vadim Seregin Qualcomm Marta Karczewicz Qualcomm
Abstract
In VTM-5.0, clipping values of adaptive loop filter are calculated with floating-point arithmetic operations. In this proposal, the calculation of clipping values is simplified by using integer operations. Compared to VTM-5.0, the proposed methods achieve similar coding efficiency and encoding/decoding time under common test conditions.
Meeting Notes
JVET-O0532 Non-CE5: Simplification on clipping value calculation for adaptive loop filter [N. Hu, V. Seregin, M. Karczewicz (Qualcomm)]

In VTM-5.0, clipping values of adaptive loop filter are calculated with floating-point arithmetic operations. In this proposal, the calculation of clipping values is simplified by using integer operations. Compared to VTM-5.0, the proposed methods achieve similar coding efficiency and encoding/decoding time under common test conditions.

In the discussion, it is agreed that the definition that requires a floating point operation is not desirable and may be ambiguous. It is suggested to use a table definition instead, which for bit depths 8,9,10,…16 times the 4 clipping indices specifies the rounded integer output (4x9=36 entries). Same table should be used for luma and chroma, and this would be the luma formula

filterClips[ sfIdx ][ j ] = Round( 2( BitDepthY * ( 4 − alf_luma_clip_idx[ sfIdx ][ j ] ) / 4 ) ).

Cited:
JVET-O0983 X JVET-P0911 X
ARCHIVE :
JVET-O0532-v1.docx Test1.xlsm Test2.xlsm Test3.xlsm
Decisions
filterClips[ sfIdx ][ j ] = Round( 2( BitDepthY * ( 4 − alf_luma_clip_idx[ sfIdx ][ j ] ) / 4 ) ).
Citation