Organisation internationale de normalisation

4Scalable video coding (SVC)

4.1SVC bit depth and chroma format scalability

4.1.1.1.1JVT-Z036 ( Prop Reqs) [A. Segall (Sharp)] On the requirements for bit-depth and chroma format scalability

This document discusses the requirements for bit-depth scalability within the context of consumer applications. Current trends in display technology are the focus, and it is asserted that these trends motivate the need for higher bit-depth formats within consumer devices. Thus, it is proposed that development of any bit-depth scalable system should consider these applications.
The contribution contiains an emphasis on larger visual dynamic range, as opposed to increased precision representation of the same visual dynamic range (adding least significant bits).
The contribution suggests that 10 bit sample depth support is necessary for near term; 12 bits for longer term.
An approximate 10¹⁴:1 human visual dynamic range was reported; 10⁴:1 in a short time interval. CRTs can do that, but with ambient light this is reduced to 50:1 or 100:1.
Multiple displays are on the market – getting brighter and extending their dynamic range, with very wide contrast ratios emerging. Example companies producing such technology: Sharp, LG, Brightside/Dolby.
An HDR image was demonstrated.
Remark: Justifying high bit depth and justifying scalable support for it are somewhat different subjects. Our previous work has already included increased bit depth support (except in the scalability context).
JVT conclusion: It seems generally agreed that support of a large visual dynamic range is an important capability to deliver in our work.

4.1.1.1.2JVT-Z039 ( Info) [S. Liu, A. Vetro (MERL)] Requirements for bit-depth scalable coding

This contribution considers a new application scenario for bit-depth scalable coding in which receiver-side editing of a high dynamic range video is desired. Requirements for bit-depth scalable coding are described, and preliminary results that aim to demonstrate the benefits of higher-bit depth video at the receiver are shown.
The contribution contains an emphasis on editing, tone mapping, etc., with "default" representation in the base layer and having an enhancement layer to provide greater quality.
Professional, "pro-sumer", and high-end consumer applications were emphasized.
The contribution suggests to consider high bit depth scalability support and to study the benefits of post-capture editing of HDR video.
Remark: Again, there's a distinction between the need for high quality / high bit depth, and the need for scalability, which is something else.
Response: There is the argument for an "easy to access" default representation supplemented by extra enhancement information – consumer quality plus extra data for high-quality subsequent processing.

4.1.1.1.3JVT-Z045-Q (Late Prop 2.2/3.1) [Y. Yu, S. Gordon, M. Yang (Broadcom)] Bit depth SVC with a prediction filter

Supporting presentation of the late contribution:

• 
  P. Topiwala
  
• 
  A. Vetro
  
• 
  Y. Chiu
  
• 
  A. Segall
  

Apology? Has been sent.

This contribution describes research work on bit depth SVC. By applying a filter to the reconstructed image from the lower layer, an average of 4.4% BDBR improvement, or an average of 0.15 dB BDPSNR improvement, can reportedly be achieved at the 10 bit top layer for "Viper" sequences. Higher gain is reportedly seen on input sequences with normal lighting conditions. (One sequence had a reported benefit exceeding 10%.)
The contribution describes the use of bit-depth scalability with tone mapping as in the current JSVM.
The contribution proposes modification of tone mapping operation.
Had some problem with BD measurements.
A presentation deck and "r1" version of the document were uploaded later.
Question: Has application of the filter after adding the enhancement layer been tried? Response: No. Remark: That might provide equivalent gain (without adding mandatory complexity to the decoding process).
Remark: Better gains in daytime scenes – good lighting conditions. Response: That's right. Remark: The sequences were intentionally chosen to provide a mixture of daytime and night-time content.
Further study is encouraged, including consideration of the post-processing comparison question.