The Joint Collaborative Team on Video Coding (jct-vc) of itu-t wp3/16 and iso/iec jtc 1/ sc 29/ wg 11 held its tenth meeting during 11-20 July 2012 at the City Conference Centre (ccc), a k. a

5.7.3Interlaced scan video coding (2)

JCTVC-J0466 Performance of HEVC for Interlaced Video [A. Luthra, D. Baylon (Motorola Mobility)] [late]

The performance of HEVC for interlaced scanned video is characterized. For the interlaced video sequences tested, the results reportedly show that when compressed in field mode the performance of HEVC can significantly degrade to require as much as 21% more (luma) bits in comparison to AVC for the sequences with less motion. Alternatively, if the interlaced video sequences are compressed in frame mode, the performance of HEVC can reportedly significantly degrade to require as much as 24% more (luma) bits in comparison to AVC for other sequences which have large motion. For some sequences the change in the coding efficiency (number of bits required) of HEVC can reportedly be as large as 60%, depending upon frame or field selection. It is noted that in HEVC it is not possible to, for example, code an intra field followed by a predicted field followed by a predicted frame. As the motion characteristics in many sequences can change with time, the inability to adapt the frame-field decision in HEVC at the level below the sequence level will therefore impact the performance of HEVC in comparison to AVC for interlaced video.

Several tables of information were provided. Some of the data, showing luma results versus adaptive frame/field coding for AVC, are shown below. All-frame coding was also studied.

Sequence	HEVC field vs. AVC	HEVC frame vs. AVC	Best
Mobile and Calendar	21.34%	‒22.44%	‒22.44%
Trapeze	‒29.81%	‒8.12%	‒29.81%
Tempete	5.96%	‒25.39%	‒25.39%
F1car	‒35.32%	13.11%	‒35.32%
Stefan	‒24.53%	23.13%	‒24.53%
Hockey	‒51.10%	24.61%	‒51.10%
Tennis	‒16.11%	‒12.46%	‒16.11%

Tennis and Trapeze were suggested to be particularly interesting in terms of mixed-motion characteristics.

One suggestion was to consider comparing the better of all-frame HEVC and all-field HEVC in each case. With that type of comparison, there would seem to always be a substantial bit rate savings for using HEVC. This is shown in the added column on the right in the above table. GOP-by-GOP optimization would probably do even better (although open-GOP would not be possible at the transition points).

A participant asked whether the configuration files used in this test could be made available.