Of itu-t sg16 wp3 and iso/iec jtc1/SC29/WG11

JCTVC-M0427 AHG8: Cross-check of residue rotation and significance map context for screen content coding (JCTVC-M0333) [J. Wang, D. He (??)] (Blackberry)] [late]

JCTVC-M0350 AHG8: Video coding using Intra motion compensation [M. Budagavi, D.-K. Kwon (TI)]

(Discussion chaired by D. Flynn.)

Applications such as wireless displays, automotive infotainment, remote desktop, remote gaming, cloud computing etc. are becoming popular. Video in these applications often has mixed content consisting of natural video, text, graphics etc. In text and graphics regions, patterns (e.g. text characters, icons, lines etc.) can repeat within a picture. This contribution proposes a CU-level intra motion compensation tool to remove this redundancy to reportedly achieve coding gain. When intra motion compensation is enabled for a CU, either horizontal motion or vertical motion is signalled. The proposed method is tested with two different search ranges in the encoder. In the first test, horizontal and vertical motions are limited to a range of 0 ~ −63. In the second test, the vertical motion is further limited so that displaced block does not go beyond LCU boundary. The second test is also combined with Sample adaptive angular Intra prediction in horizontal and vertical direction only (SAPHV of JCTVC-M0056) and coefficient level cRiceParam update of JCTVC-M0316. The following bit rate savings under RCE2 common conditions (lossless coding) and on RCE2 screen content sequences (RGB 4:4:4) are reported:

• 
  First test: Intra MC horizontal only or vertical only, Search range of 0 ~ −63: AI: 24.0%, RA: 18.9% and LDB: 15.7%.
  
• 
  Second test: Intra MC horizontal only or vertical only, Search range of 0 ~ −63, Vertical motion limited to be within LCU: AI: 21.0%, RA: 16.6% and LDB: 14.6%.
  
• 
  Second test + Sample adaptive angular Intra prediction in horizontal and vertical direction only (SAPHV) of JCTVC-M0056: AI: 29.1, RA: 22.6 and LDB: 19.8%.
  
• 
  Second test + SAPHV + cRiceParam update (JCTVC-M0316): AI: 32.8, RA: 25.9 and LDB: 22.9%.
  

Further improvements to this technique including extensions to handle 4:2:0 and 4:2:2 content are currently being studied.

It should be noted that the term motion compensation does not have the conventional interpretation of representing motion, rather this is a block copying operation within a picture. A more appropriate term would be desirable to avoid confusion.

The block-copying operation is performed on integer pels and is restricted to a 1D search in either the horizontal or vertical directions. The block-copying mode transmits a single displacement relative to the current CU position and covers the entire "2Nx2N" intra CU, i.e., despite the name, the PU structure of inter prediction is not reused. Therefore the minimum prediction block size is 8x8 for intra.

To reduce line buffering requirements, a variant is presented that constrains the vertical displacement to only reference data within the current LCU, yielding a ~3% performance loss.

It is demonstrated that, when used in conjunction with SAPHV, a further 8% performance improvement can reportedly be achieved, suggesting that the mode is not in conflict with this tool.

This reportedly is a simple initial implementation of such a tool, inspired by the dictionary coding techniques presented at previous meetings. Entropy coding for this tool has not been extensively investigated. It currently codes a single bit dimension selector (horizontal or vertical) and a seven bit displacement, both bypass coded.

When compared against the dictionary coding technique of L0303, under the same common conditions, the dictionary coder reportedly achieves a 10% average performance improvement over the method 1 results of this tool in all-intra configurations.

Further study is suggested, perhaps in the context of an AHG to evaluate the merits of such a technique against that of dictionary coding.