International organisation for standardisation organisation internationale de normalisation

18.8Intra prediction

18.8.1.1.1.1.1.1.1JCTVC-C040 Chroma intra prediction using reconstructed luma [C. Yeo, Y. H. Tan, Z. Li, S. Rahardja (I2R)]

Intra coding in the current TMuC achieves high compression efficiency, in part due to the intra prediction process that exploits spatial directional correlation. However, intra prediction of chroma components in YUV 4:2:0 video uses a limited set of possible predictions available for coding of luma components. Also, the coding of chroma components proceeds somewhat independently of luma components, and ignores any possible correlation between them. This contribution proposed a way of using reconstructed luma pixels to help with intra prediction of chroma pixels. Specifically, it uses the regions with similar luma texture characteristics to identify a location for prediction of chroma components. By making use of the reconstructed co-located luma block to perform template matching in the luma plane, co-located chroma blocks of the matched luma blocks are identified for us as predictors. Simulations results reportedly indicate that, on average, the proposed approach is able to obtain 2.3% chroma bit-rate reduction and 0.5% luma bit-rate reduction relative to the anchor for all intra coding.

An approximate 2x encoding time and 3x decoding time impact was reported for intra.

It was remarked that something like this that links together luma and chroma operations would need to provide a very good gain to be a desirable feature.

The complexity seemed rather high, but further study was encouraged.

18.8.1.1.1.1.1.1.2JCTVC-C206 Chroma intra prediction by scaled luma samples using integer operations [J. Chen, V. Seregin (Samsung)]

This contribution proposed an integerization of the algorithm presented in JCTVC-B021 in which chroma samples are predicted from colocated reconstructed luma samples by a linear model relationship. Only addition and multiplication operations are involved to calculate the model parameter instead of using divisions as in the proposed approach. Experiments reportedly show that the average BD-rate reduction of Y, Cb, Cr components are 1.3%, 6.8% and 5.6% for intra coding conditions and 0.7%, 6.7% and 4.6% for random access conditions. The proponent asserted that the complexity of the proposed method is similar or even smaller than that of the method used in AVC.

The values of the parameters of the linear prediction model are derived at the decoder side from neighbouring reconstructed samples.

Reportedly the late document JCTVC-C314 contained a cross-verification.

Further study was encouraged.

18.8.1.1.1.1.1.1.3JCTVC-C314 Verification results of Samsung's proposal JCTVC-C206 [J. Kim , J. Park , B. Jeon] (late reg.)

The contents of this late document were verbally reported during the presentation of JCTVC-C206. The document was not reviewed in detail. The PSNR and bit-rate results reported in this document were reported to be the same as the results reported in the document JCTVC-C206.

18.8.1.1.1.1.1.1.4JCTVC-C111 Integration of plane mode in unified intra prediction [G. Li (Santa Clara Univ.), L. Liu, N. Ling (Santa Clara Univ.), J. Zheng, P. Zhang (Hisilicon)]

The current default intra prediction method in the TMuC, Unified Intra Prediction, reportedly has up to 33 directional modes and 1 non-directional mode, which is the intra DC mode. This contribution proposed adding a plane mode and using an "multiple predictor set" intra mode prediction. Experimental results reportedly showed that coding efficiency can be improved with the proposed method.

Partial results for intra HE and LC configurations were provided. An average gain of about 0.6% was reported.

There seems to be several elements discussed together in this contribution.

The encoder search complexity is increased.

It was noted that the encoder search is limited by using fast search in the current TMuC, and better compression can be obtained just by making a more effective search.

Further study was encouraged.

18.8.1.1.1.1.1.1.5JCTVC-C189 Evaluation of intra prediction based on repetitive pixel replenishment [S. Mochizuki, K. Iwata (Renesas)]

This contribution proposed a method of intra prediction based on repetitive pixel replenishment (Intra RPR).

The proposal uses a displacement vector with area extrapolation for intra coding.

Preliminary partial simulation results reportedly showed that this proposed technique has an average of 0.8%, 0.2%, 1.3% BD-rate improvement, respectively at Y, U and V, against TMuC 0.1, with about double encoding time. Only the encoding the first 20 pictures of each sequence was tested.

The reported maximum savings per sequence was 4.8% (0.22 dB) on the Basketball drill relative to TMuC 0.1. The contributor indicated that they are working on integration into a newer TMuC version.

The design, as presented, was somewhat preliminary – some modifications were planned (making the prediction size adaptive, and harmonization with other intra coding tools).

Further study was encouraged.

18.8.1.1.1.1.1.1.6JCTVC-C207 Encoder improvement of unified intra prediction [Y. Piao, J.-H. Min, J. Chen (Samsung)]

This contribution proposed an encoding scheme for unified directional intra prediction (UIP) which was introduced in JCTVC-B100.

This was only a non-normative proposal – only the encoder would be changed – not the decoder.

Bit rate reductions of 0.5% for HE all-intra and 0.3% for LC all-intra were reported relative to the TMuC method, with lower search complexity.

The results were not cross-verified.

Two variants were proposed.

Decision: The S1 (9-9-4-4-5) scheme was adopted as the HM encoder method, with additional further study planned, and the S2 scheme will additionally be available in the HM software as a selectable configuration.

18.8.1.1.1.1.1.1.7JCTVC-C218 Encoding complexity reduction for intra prediction by disabling NxN partition [J. Kim, Y. Jeon, B. Jeon (LG Electronics)]

This contribution proposed a simplification of the intra prediction process to reduce encoding complexity. In the current TMuC specification, two kinds of splitting, 2Nx2N and NxN, are allowed for intra prediction and it is signaled by intra_split_flag. In the proposed method, NxN is not allowed for each intra prediction unit (PU) except when the PU has the maximum depth. The proposed modification reportedly achieves approximately 4.5% encoding time reduction with 0.2% loss in BD bit rate under the intra HE setting of JCTVC-B300. When the rotational transform (ROT) and combined intra transform (CIP) are turned off, the proposed method reportedly achieves about 36% encoding time reduction with 0.4% loss in BD bit rate relative to the anchor.

It was noted that this proposal includes syntax modification.

It was remarked that the software includes a macro switch for a similar treatment in the inter case.

Further study was suggested – perhaps as part of block structure testing rather than intra prediction.

18.8.1.1.1.1.1.1.8JCTVC-C234 Simplified intra smoothing [Y. Zheng, M. Coban, M. Karczewicz (Qualcomm)]

In this contribution, a simplified intra smoothing scheme was proposed to improve the current AIS scheme. In the proposed scheme, the prediction samples are filtered according to the prediction unit size and the intra prediction mode. The proposed scheme reportedly achieves 0.4% BD bit rate reduction on average for HE and LC intra configurations, relative to not using any intra smoothing. The encoding complexity was reportedly reduced by about 50% relative to the prior AIS scheme.

No syntax is used to indicate whether to filter or not.

Visual quality was not studied.

Random access and low delay configurations were not tested.

The late document JCTVC-C316 reportedly contained a cross-verification report.

Further study was encouraged.

18.8.1.1.1.1.1.1.9JCTVC-C316 Verification Results of Simplified Intra Smoothing (JCTVC-C234) [T. Yamakage (Toshiba)]

The contents of this late document were verbally reported during the presentation of JCTVC-C234. This contribution reported verification results of JCTVC-C234 (Simplified Intra Smoothing) by Qualcomm. Not all results had not been checked, but the reported results were indicated to be similar to Qualcomm’s results.

18.8.1.1.1.1.1.1.10JCTVC-C302 Improved intra smoothing for UDI and new AIS fast mode [T. Nguyen, B. Bross, D. Marpe, T. Wiegand (HHI)] (late registration, missing prior, available first day)

In this contribution, an adaption of an intra smoothing scheme for UDI was proposed to improve the overall coding performance. Furthermore, this contribution presented modified fast mode for AIS.

For when AIS is not in use, the contribution proposed a LUT without syntax to determine when filtering is to be applied. For when AIS is active, the contribution proposed a syntax indicator using the table as part of the entropy coding, and a fast encoding search method to guide the selection.

Relative to not using filtering, the LUT-only scheme in the proposal was reported to provide about 0.3% for HE and LC all-intra.

Relative to not using filtering, the scheme using syntax in the proposal was reported to provide about 0.8% for HE and 0.9% for LC in all-intra coding.

It was suggested that visual quality effects would be desirable to assess.

The contributor did not suggest to put this into the TM, but rather to do further study.

18.8.1.1.1.1.1.1.11JCTVC-C269 A study on the impact of intra smoothing [A. Tabatabai, C. Auyeung, T. Suzuki (Sony)]

Please see the discussion of this contribution in section 15.2.