International organisation for standardisation

6.4Functionalities

6.4.1Scalable coding

See also notes of joint meeting with MPEG and VCEG as recorded below.

Note: Some contributions related to temporal scalability (e.g. JTVC-F462 and JCTVC-F546) are discussed in the section on decoded picture buffering.

6.4.1.1.1.1.1.1.1JCTVC-F290 Scalability Support in HEVC [D. Hong, W. Jang, J. Boyce, A. Abbas (Vidyo)]

This contribution proposes a scalability extension to the HEVC design. It applies to both intra and inter coding. The design attempts to minimize changes from the enhancement layer codec from the single layer HEVC codec, re-using existing tools exactly wherever possible. Experiments were performed for 2-layer spatial scalability with a 1:2 ratio in each dimension, although the method may be used for any resolution ratio between the layers, including no resolution change for CGS, or different scaling factors in the two dimensions. The contribution reports BD-rate gains vs. simulcast of 17.9% for Intra HE, 16.3% for Intra LC, 16.1% for RA HE, 15.5% for RA LC, 11.4% for LD HE, and 11.1% for LD LC. The proposed scalability extension may also be used with a mixed codec design, with an H.264/AVC base layer and scalable HEVC-based enhancement layer.

Method reportedly similar to SVC: Enhancement layer in “Difference mode” with base-layer upsampling (using DCTIF) and “sample mode” (without base-layer referencing). Gains were reportedly similar as for use of the JSVM for intra and RA.

Contribution noted (information about reported method with some aspects which are outside the scope of the current technical work of JCT-VC).

High layer syntax must be constructed such that it is extensible. It was suggested that the first byte of the NAL unit should be kept unchanged in any extension.

6.4.1.1.1.1.1.1.2JCTVC-F292 Metrics for evaluation of scalable coding [J. Boyce (Vidyo)] [late upload 07-07]

The contribution provided input to the definition of test and evaluation conditions for the planned scalability extension to HEVC, for coding efficiency comparison to simulcast and single layer coding. The same raw experimental data provided in the JCTVC-F290 contribution is presented using three different analysis methods. The contribution recommended one analysis method, in which the simulcast high resolution and scalable enhancement layer Y-PSNR values are matched.

In general, it would be desirable to know how scalable coding performs in comparison versus single layer and simulcast. For proper comparison, it is suggested that the base layer should be the same as simulcast.

6.4.1.1.1.1.1.1.3JCTVC-F096 Scalable structures and inter-layer predictions for HEVC scalable extension [H. M. Choi, J. Nam, D. Sim (Kwangwoon Univ.)]

This contribution reports an investigation about two kinds of scalable structures and inter-layer predictions for a HEVC scalable extension. Two kinds of structures are single- and multi-loop designs. In the contribution, two different inter-layer predictions which can be applied to two scalable structures are proposed. An inter-layer texture prediction (ILTP) is a scheme to predict PU of the enhancement layer from the corresponding block of a reference layer. For the other inter-layer prediction, a generalized inter-layer reference frame (GILR) will be presented. In this contribution, each scalable structure is evaluated with two inter-layer predictions. The single-loop scalable structure with ILTP gives on average BD bit rate reduction about 14% in all intra case for luma signals, compared to simulcast only for the enhancement layer. The multi-loop scalable structure with ILTP also gives the similar RD performance. For the single- and multi-loop scalable structures with the GILR, a coding gain of approximately 20% is reportedly observed on average in BD bitrate for all-intra case, compared to simulcast for the enhancement layer.

Contribution noted (information about reported method with some aspects which are outside the scope of the current technical work of JCT-VC).

6.4.1.1.1.1.1.1.4JCTVC-F618 Resampling filters for scalability and screen content applications [W. Dai, M. Krishnan, P. Topiwala (FastVDO)]

Discussed in Track B: Alternative coding modes

Spatial resampling filters are in wide use in image and video processing applications, as there is often a mismatch between the native resolution of coded video data and that of target displays. Spatial resampling is even used within the context of video coding itself, such as in spatial scalability, where a mixed representation requiring multiple resolutions is an explicit part of the coding architecture. In SVC for example, resampling filters are standardized (more specifically, the upsampling filter, needed in the decoder, is standardized). Note that no matter how well designed, the use of resampling filters nearly always involves a loss of information (when applied in contexts other than perfect reconstruction filter banks), which one naturally wishes to minimize. Resampling filters are also regularly used with little notice for color space representations, as most video codecs take YUV 4:2:0 as input, whereas the native representation of acquired video is typically RGB (4:4:4). Similarly to spatial resampling, there is information lost in the color space resampling to YUV 4:2:0. In fact, for certain applications, working in fully 4:4:4 may be justified. In the development of HEVC, 4:4:4 coding is already part of the mandate, although progress in this area has yet to begin. The use of 4:4:4 and 4:2:0 has recently emerged as a possible issue on the reflector for the coding of screen content video in HEVC. Likewise, scalability is also of current interest in the development of HEVC and future potential extensions. In this proposal, spatial filters are described which can potentially provide improved quality for both spatial resampling, as well as chroma sampling.

The contribution was just for information purposes. It was currently not to be considered, as it is not required to specify normative upsampling filters. In general, only one side needs to be considered as normative (typically the upsampling filter).

6.4.1.1.1.1.1.1.5JCTVC-F488 Requirements for Scalable extension of HEVC [E. François, S. Lasserre, F. Le Leannec (Canon)]

This contribution aims at proposing requirements in the context of the ad-hoc group on requirements for the scalable extension of HEVC, established in March 2011. The intent is to bring inputs related to mobile and video capture devices. In particular, requirements related to complexity are discussed.