IRR/IU/vid/de-interlacing

IRR IU Task Group
rev. Tue Nov 9 17:13:25 EST 2004
Steven D. Johnson

D-interlacing in xine - Steve Johnson

I have been experimenting with xine options for de-interlacing in the hopes of improving the frames captured from the DVD clips. De-interlacing is activated with the -D option to the xine command. The software release includes several methods.

The default configuration file (~/.xine/config) generated on my desktop includes the line:

  #gui.deinterlace_plugin:tvtime:method=LinearBlend,\
                                        cheap_mode=1,\
                                        pulldown=0,\
                                        use_progressive_frame_flag=1

In each test, I added the parameter chroma_filter=1. According to commentary in xine source file src/post/deinterlace/xine_plugin.c:

Chroma_filter: DVD/MPEG2 use an interlaced image format that has a very poor vertical chroma resolution. Upsampling the chroma for purposes of deinterlacing may cause some artifacts to occur (eg. color stripes). Use this option to blur the chroma vertically after deinterlacing to remove the artifacts. Warning: cpu intensive.

So a representative innvocation looks like:

xine -Dtvtime:method=LinearBlend,chroma_filter=1,pulldown=0,use_progressive_frame_flag=1 VOB file

I have not yet varied the other parameters, such as pulldown to see the effect.

none 1 2 3 4 5 6 7 8 9
Default (-D) 1 2 3 4 5 6 7 8 9
LinearBlend 1 2 3 4 5 6 7 8 9
Greedy2Frame 1 2 3 4 5 6 7 8 9
ScalarBob 1 2 3 4 5 6 7 8
Vertical 1 2 3 4 5 6 7 8 9
Weave 1 2 3 5 7 8 9

Observations (so far):

Generally, enabling deinterlacing does seem to improve the resolution somewhat.
The LinearBlend method seems best, but there are other parameter settings yet to be explored.
There are significant artifacts, listed below. I don't think these are attributable to the .PNG format. In order to see these, you may need use a viewer like xv and magnify. The images are at
/l/ddd/irrt/img/dvd3/xineD/Method_Name/*.png
- In some frames, features like towers appear "double exposed". See [frame] for a typical example; a few instances are much worse. I suspect that this effect is a consequence of camera vibration, and so it tells us that vibration will be a significant issue.
  (detail)
- In most frames, there are vertical "banding" artifacts as noted in the source comment above. See [frame] for a typical example. Bushes are a problem. Even with chroma_filter enabled (=1, which I think means enabled) there are bands, but the effect appears usually to be diminished.
  (detail)
- At the same time, chroma_filter also seems to reduce the over-all resolution, but I'm not sure this is a bad thing in contrast to the staggering caused by interlacing.

This is on-going experimentation, and I'll update this index as I learn more. I hope that we won't need these grainy .VOB files too much longer, but right now they are the closest thing we have to representative footage. I hope to use xine for visualization and instrumentation later on.

none	1	2	3	4	5	6	7	8	9
Default (-D)	1	2	3	4	5	6	7	8	9
LinearBlend	1	2	3	4	5	6	7	8	9
Greedy2Frame	1	2	3	4	5	6	7	8	9
ScalarBob	1	2	3	4	5	6	7	8
Vertical	1	2	3	4	5	6	7	8	9
Weave	1	2	3		5		7	8	9