Back to Search Document details
15th Meeting: Gothenburg, SE, July 2019 2019-07-05 10:26
Non-CE5: On Exp-Golomb Coding of ALF Filter Coefficients and Clip Values
Abstract
This document presents various simplifications for signaling of Exp-Golomb order (EGk) used to build codes for signaling ALF filter coefficients and clip values. In VTM-5.0, the EGk signalling involves a recursive prediction among neighboring EGks and an additional flag used as an offset for the prediction. In this contribution, all presented simplifications remove the recursive prediction and the offset flag. Further tested simplifications include (i) signaling a single EGk for all positions instead of signalling separate EGks for different groups of positions, (ii) and using a fixed EGk without signalling.
JVET-O0648 Non-CE5: On Exp-Golomb Coding of ALF Filter Coefficients and Clip Values [H. E. Egilmez, N. Hu, V. Seregin, M. Karczewicz (Qualcomm)]

This document presents various simplifications for signalling of Exp-Golomb order (EGk) used to build codes for signalling ALF filter coefficients and clip values. In VTM-5.0, the EGk signalling involves a recursive prediction among neighbouring EGks and an additional flag used as an offset for the prediction. In this contribution, all presented simplifications remove the recursive prediction and the offset flag. Further tested simplifications include (i) signalling a single EGk for all positions instead of signalling separate EGks for different groups of positions, (ii) and using a fixed EGk without signalling.

The experimental results under common test conditions show that the proposed simplifications lead to almost no performance change.

The results demonstrate (see slide 7 in ppt of upload v4) that along with the adoption of JVET-O0064 for fixed length coding of clip values the version with EG3 for coefficient coding (and without prediction as per adoption of JVET-O0669) has almost identical results as EGk, but is simpler.

Decisions
adopted
The results demonstrate (see slide 7 in ppt of upload v4) that along with the adoption of JVET-O0064 for fixed length coding of clip values the version with EG3 for coefficient coding (and without prediction as per adoption of JVET-O0669) has almost identical results as EGk, but is simpler.
Citation