Back to Search Document details
4th Meeting: Chengdu, October 2016 2016-10-25 11:22
Bilateral filter after inverse transform
Authors: Jacob Ström Per Wennersten Ying Wang Kenneth Andersson Jonatan Samuelsson
Abstract
This contribution presents a bilateral filter operation, suggested to be performed on decoded sample values directly after the inverse transform. The proposed bilateral filter is a five-tap filter in the shape of a plus sign. The strength of the filter is based only on the TU size and QP. No additional parameters are determined during encoding and no new syntax elements are proposed except for flag at slice-, PPS- or SPS-level for turning on or off the filter. It is reported that the overall BD-rate improvement for JVET sequences is 0.5% for AI. It is also reported that the overall BD-rate improvement for JVET sequences is 0.50% Y for RA. A complexity increase of about +9% / +7% is reportedly estimated for encoder/decoder for AI and about 4%/3% is estimated for encoder/decoder for RA. It is proposed to create an exploration experiment to study the bilateral filter in detail until the next JVET meeting.
Meeting Notes
JVET-D0069 Bilateral filter after inverse transform [J. Ström, P. Wennersten, Y. Wang, K. Andersson, J. Samuelsson (Ericsson)]

This contribution presents a bilateral filter operation, suggested to be performed on decoded sample values directly after the inverse transform. The proposed bilateral filter is a five-tap filter in the shape of a plus sign. The strength of the filter is based only on the TU size and QP. No additional parameters are determined during encoding and no new syntax elements are proposed except for flag at slice-, PPS- or SPS-level for turning on or off the filter. It is reported that the overall BD-rate improvement for JVET sequences is 0.5% for AI. Preliminary results reportedly indicate that the overall BD-rate improvement for JVET sequences is 0.50% Y for RA. A complexity increase of about +9% / +7% is estimated for encoder/decoder for AI and about 4%/3% is estimated for encoder/decoder for RA. The contribution proposed to create an exploration experiment to study the bilateral filter in detail until the next JVET meeting.

The filter strength is dependent on QP.

LUT operation makes complexity manageable.

RA results were not ready yet, but were expected to be similar.

Investigation in an EE was planned.

It was asked whether it would be better to perform the filtering not in the intra prediction loop, but rather before deblocking. Proponents reported that such configuration was tested but had somewhat lower coding gain. However, the current block-wise interleaving of processing with intra prediction might be undesirable. Therefore, such an alternative configuration should also be investigated in the EE.

Cited:
JVET-D0001 JVET-D0184 X JVET-E0001 JVET-E0010 JVET-E0031 JVET-E0032 JVET-K0068 JVET-K0231 JVET-K0274 JVET-L0172 JVET-M1021
ARCHIVE v8 :
JEM3-vs-D0069_paralell_classA.xlsm JVET-D0069_Bilateral_Filter_After_Inverse_Transform_r5.docx JVET-D0069_Bilateral_Filtering_After_Inverse_Transform_r5.pptx
Decisions
It was asked whether it would be better to perform the filtering not in the intra prediction loop, but rather before deblocking. Proponents reported that such configuration was tested but had somewhat lower coding gain. However, the current block-wise interleaving of processing with intra prediction might be undesirable. Therefore, such an alternative configuration should also be investigated in the EE.
Citation