International organisation for standardisation

6.7.3Weighted prediction and illumination compensation

6.7.3.1.1.1.1.1.1JCTVC-F265 Weighted Prediction [P. Bordes (Technicolor)]

This document presents results of Weighted Prediction (WP) and an associated analysis tool implementation into HM. This report is an AHG18 outcome.

WP has been designed to compensate illumination variation in video sequences. It is part of the AVC standard and this feature is useful in video encoder and video splicing applications in particular.

Experimental results of WP and the analysis tool have been made on fade sequences generated with a fading tool provided in AHG18.

It was reported that WP in explicit mode has a gain in the range of 17-28% (depending on the configuration) for the test sequences with linear fade when weights are known in advance, and in the range of 15-29% (depending on the configuration) when the WP analysis tool is used.

The difference from AVC in the chroma is to center around 128 rather than 0 and make adjustments for precision of the computations; otherwise the proposal is essentially the same as in AVC.

This contribution focuses on explicit prediction, and there was a prior contribution to the Geneva meeting that focused on implicit prediction (with test results).

Text and software were provided in the contribution.

The provided software did not at first appear to be sufficiently "clean".

The provided text seems to be somewhat disconnected from a contextual relationship with the draft standard – just being a separate section without connections to the remainder of the document.

The proponent suggested using software from JCTVC-F326 (software to be uploaded) rather than this contribution.

It was remarked that the high-precision averaging may not be implemented correctly in the software.

The weights for explicit weighting were placed syntactically at the slice header level (as in AVC).

It was remarked that the slice parameter set concept may be useful for sending the weights.

Text and software seem adequate to enable integration, although some further improvement of the software is needed from the proponent. Decision: Adopted.

6.7.3.1.1.1.1.1.2JCTVC-F436 AHG18: Cross-check report of Weighted Prediction, proposal JCTVC-F265 [R. Boitard, L. Guillo (INRIA)]

6.7.3.1.1.1.1.1.3JCTVC-F331 Cross check report of Technicolor's proposal of JCTVC-F265 for Weighted Prediction from Toshiba A. Tanizawa (Toshiba) [A. Tanizawa (Toshiba)]
6.7.3.1.1.1.1.1.4JCTVC-F326 Explicit Weighted Prediction with simple WP parameter estimation [A. Tanizawa, T. Chujoh, T. Yamakage (Toshiba]

This document presents experimental results for Weighted Prediction (WP) with a simple WP parameter estimation scheme. WP has been designed to compensate illumination variation in video sequences and included in the AVC standard. This feature is effective for video splicing process used in video authoring system and can get significant coding gain for fading sequences.

In this document, a WP parameter estimation method based on alpha-blending model is discussed in order to derive WP parameters under 1-pass encoding conditions. This method uses only spatial image characteristics and its computational cost is asserted to be negligible. The experimental results in HM software version 3.0 under common test conditions are reported. The WP using 1-pass encoding scheme has a large reported gain of 17% to 31% on average (up to 61% for Low delay P structure) for the test sequences with linear fade provided in the WP AHG. It is reported that the WP does not have a big impact on coding efficiency for the regular test sequences.

6.7.3.1.1.1.1.1.5JCTVC-F457 AHG18: Cross-check report of Explicit Weighted Prediction with simple WP parameters estimator, proposal JCTVC-F326 [P. Bordes (Technicolor)]

6.7.3.1.1.1.1.1.6JCTVC-F397 Weighted Prediction with Parameter Estimation [S. Takamura, Y. Bandoh, S. Matsuo, K. Kamikura, H. Jozawa (NTT)]

(Information document – experiment report.)

This report summarizes the performance evaluation of the weighted prediction tool using fade parameter estimation. For faded sequences in the low delay configuration, WP with given parameter (ground truth) provided 18-29% gain (Y), WP with estimated parameter provided 17-28% gain (Y). Encoding complexity was 100-128% while decoding complexity was 82-107%. The same estimation method as used in JCTVC-F326. Reference to estimation method: Hirofumi Aoki and Yoshihiro Miyamoto, "An H.264 weighted prediction parameter estimation method for fade effects in video scenes," Proc. ICIP 2008, pp. 2112-2115, Oct. 2008.

6.7.3.1.1.1.1.1.7JCTVC-F417 Pixel Based Illumination Compensation [Chan-Won Seo, Jong-Ki Han (Sejong Univ.), Jeongyeon Lim (SK telecom)]

This contribution describes a pixel based illumination compensation scheme to compensate illumination changes. The Weighted Prediction (WP) proposed at the last meeting has been designed to compensate illumination variation at the slice level and it was tested with fade sequences. The WP is useful but it may be not effective for natural sequences whose brightness varies locally. The IC scheme proposed in this contribution derives IC parameters at CU-level. The IC parameters are applied to a block or some pixels adaptively. Coding gains are reportedly about 0.2% and 0.3% for random access and low delay coding configurations, respectively.

It was noted that SAO and WP and ALF and DC coefficients also have effects that can somewhat overlapping effects.

In its present form, the proposal doesn't seem to be providing an adequate tradeoff. Some kind of further improvement would be needed to justify further consideration.

6.7.3.1.1.1.1.1.8JCTVC-F361 Cross-verification of Sejong Univ.'s proposal on pixel based illumination compensation [Y. Jeon, B. Jeon (LGE)] [late upload 07-12]