Blended Fields [illustrations & my thoughts]

[^^-+I4004+-^^]

as i've noticed currently it's hard to fing good explanation on blended fields and even harder to find nice illustrations of this phenomenon
(on the web & on this forum)

(i've found 1 att. in this forum but it got misnamed pictures....
the author thought there were "swapped" fields while in fact it was case of blended fields ( swapped fields look even worse than blended fields ) ! it was in thread : "Doom9's Forum > General > DVD2AVI > Pal DVD’s with blended fields" att. was : "screenshots.zip" by author : "nig103"..........i'm off-line so i'm not giving direct link.....also my browser doesn't save this forum's pages correctly etc. )

so i did my own tests & decided it would be good if i share this with forum
( for anyone that has doubts on how blended fields look like...as i did before this tests... )

i live in PAL country,so all captured video is from analog PAL source.......
( bt8x8 card,picvideo mjpeg,704x576... )

avi's were inspected with an .avs script lookin somethin like this:

avisource ("d:\video\raw_stuff\eurosport.avi")

complementparity()
SeparateFields()

[in my case without "complementparity" avisynth doesn't know the right order of the field-pairs so video goes back&forth.....so i used it to avoid that....]

this was then loaded to VDub.....this way "frame-count" of VDub becomes "field-count" ( bare that in mind when lookin at my screenshots! in every screenshot VD "field"(frame) counter is visible,that will help you to understand better ! )

in other words->if i have 1 second video (at 25fps) and load it with this script,there will be 50 "frames" counted by VD (as now every field is "mistaken" for a frame!)

BUT this in fact are fields ( 50 fields )!you'll see that because they have half of the regular PAL resolution (image is "squished")
also you'll see fields alternating ( along the timeline ) if you take good look!
(look for tv-logo and graphics on the screen...you'l see how 2 fields (that make 1 frame) differ!)


following sources were captured (and i'll distinct them for blended fields right now):

1-film source ( USA tv-series shot on film-> from the 70's )
an example of poor NTSC->PAL conversion ( HAS BLENDED FIELDS )
[ big THANKS goes to Swan for providing the source clip ! ]


2-manga cartoon ( captured from MTV-Germany-> analog SATV VHS recording )
also VERY poor conversion ( even on tv/vhs it's jerky motion! )
( HAS BLENDED FIELDS )

3-live-PAL-tv ( eurosport & few news channels-> analog SATV VHS recording )
( NO BLENDED FIELDS -> pure interlaced PAL without conversions )

4-film source ( USA tv-series shot on film->from the 90's )
( NO BLENDED FIELDS )


( note:VHS recording doesn't "touch" fields in any way -> it DOESN'T introduce
field swaps,field drops,blending fields.... so i treated VHS as a creditable source ! )

files will reside on my web , each is cca 500kB - 880kB .zip file with 10 (or so) screenshots
of VDub input window.....

link:

http://free-st.hinet.hr/ikostic3/video_/video.htm


(i hope i'll upload all zipped archives by tomorrow...1st two are uploading while this is being pasted to forum....)


at last i would like to recommend downloading to everyone with any interest in
interlacing stuff........
( filenames speak for themselves.....if you got problems dloading let me know (pm or email me ) and i'll try to see if anything can be done to resolve issue....in each file i'll enclose short .txt file that will try to go a bit further in the techniques involved in this kind of standards conversion....
purely unprofessional comments,but after lookin at pics perhaps you will agree with me on how blended fields are produced by hardware ("on-the-line") standards converters....... )


[ i would like to thank all members of this forum and especially Swan & neuron2 for inspiring me to do this...( in thread "music videos-interlaced question" on this forum..)....hope everyone can benefit from this
illustrations... ]


my greetings to you people.......

Ivo

[Swan]

This is awesome stuff!
Thank you so much!
Lots of useful and entertaining info to take in and digest, Ivo.
This clears up a lot of questions I've had about my captures.
Once again, thank you for this enlightening "essay". :-)

/Swan

[droolian01]

Hi there.

This topic of blended fields has been discussed quite a few times here, and i think the reason why there isn't many screenshots of it is because anyone who has suffered from this can provide thousands of examples (during my 24 caps i bet i could provide a couple of hundred examples from each episode!!!)

A quick search found ;-
http://forum.doom9.org/showthread.php?s=&threadid=24940 (http://)

By definition information is unrecoverably lost so it is imposible to do any sophisticated ivtc type operation - all you can do is try to make them less damaging.

I had an idea once to reduce this problem - but it is way to complicated for me to work out (just started chapter 2 of a C textbook i got from a charity shop - i'll be doing filters one day! [just don't hold your breath!!]} It would involve some form of interploation to create a new field, but i think that this would only produce a similar effect to blend de-interlacing methods currently available - anyway, how could you detect a frame as blended reliably? Way too much effort (if feasable) for too little (if any) return.

In short you cannot recover the lost data, all you can do is blend away the effects. This is just a fact.

p.s. Any U.S. readers here answer this one for me (i think i know the answer anyway - just nice to have it confirmed) Do you know what we are talking about???? All your tv caps should be pefectly ivtc'able (you had to get some compensation for ntsc 480 res!) the only problem i could see is with native uk/european pal tv stock (not film) but then i bet you only get frame duplication and not blending (if at all).

Interesting!

[Swan]

I had not given this much thought until recently. I capture a U.S. 70's police series on a regular basis (at 704 x 576) and have always seen "combing" and thought that's how it's supposed to be.

But then I got to thinking. "This material is clearly from a film source. Shouldn't I be able to recover the progressive frames instead of using FieldDeinterlace? That ought to give better quality on the encoded DivX."

And, I also thought "if the source is Progressive (each fields contain half of the info from one frame and match perfectly) then why do I see "combing" at all?"

I tried Telecide () and the other variants Neuron2 suggested in the thread about deinterlacing music videos (http://forum.doom9.org/showthread.php?s=&threadid=34980) and encoded an episode of my beloved Police series that way. The final video looked awful. Movement by cars was jerky, it looked just like severe frame drops. Picture was stable (not jumping up and down) but no movement by people cars, nothing looked right. Something was clearly wrong here.

The more mysterious thing was that my captured 1 hour clip contained 5 minutes of the show that ran before the one I was interested in.
It was another 70's series, but it looked perfect on the monitor! No combing! Was it possible that is was a good PAL conversion? While the show I wanted, was not? I could hardly believe my eyes, there was no combing at all.

This is where Ivo came in. Thank you so much; your post explains it all to me. And thank you too, Droolian. I understand now there's nothing I can do about it, I have to use FieldDeinterlace ().

How often do you ever capture anything that's been correctly converted like this:
Film --> NTSC
NTSC-->Back to Film-->PAL
Almost eveything I capture appear to be:
Film --> NTSC --> PAL (blended fields problem)

And is it possible to easily tell the difference between blended fields made by a poor NTSC-->PAL conversion and progressive PAL with a simple field shift? Both look interlaced (combed) on the monitor.
Or does one have to create an Avisynth script for each captured video using SeparateFields() and look at each field to see if they're blended to determine this?

What causes a simple field shift?

/Swan

PS This is one of my favorite threads on the subject:
http://forum.doom9.org/showthread.php?s=&threadid=20032&highlight=decomb+plugin

[^^-+I4004+-^^]

d>This topic of blended fields has been discussed quite a few times here, and i think the reason why there isn't many screenshots of it is because anyone who has suffered from this can provide thousands of examples (during my 24 caps i bet i could provide a couple of hundred examples from each episode!!!)


_that's poor excuse.......i didn't saw any good screenshots....as i've said there were few,but misnamed.....so what are they good for?
(if one is a beginner?)
also you dont have to post ALL blended fields screenshots,do you?


d>!!]} It would involve some form of interploation to create a new field, but i think that this would only produce a similar effect to blend de-interlacing methods currently available

_this idea is good:as far as i know pro-grade standards converters do it simillary (if they are any good->5 field memory etc.!)
they create "new" (interpolated) fields in the right places...
these fields are "averages" of 1 field before & one after....
de-interl. works on per-FRAME basis,so it has nothing to do here--->
problem should be adressed on per-FIELD basis!
(my explanation in "manga" zip file on how blended fields are produced concerns the crude and primitive way....stuff before digital image processing...when all had to be analog..i guess newer (BAD ones) converters take 2 fields,merge them (into one frame),blend deinterlace and then take field out of that..that would explain "blended" look ..... or they just blend directly on fields....anyhow this destroys the picture..)

d> anyway, how could you detect a frame as blended reliably?

_you mean in the filter itself?
you've made a mistake->in this case you cannot say "blended frame"..we're talking fields!
you have to work only on fields!
i guess detection can be made because these fields have parts of 2 other fields->so filter should do this:take 3 (or more) fields

then it looks for "suspicious" (suspicious has half the information of the both frames therefore half the contrast or the light in the "moving" areas etc.)
suspicious one has the parts of the both fields..if it detects that (blending of fields) then it would throw away that blended one and make new one->one that is interpolation between those two fields that are good!
i agree that the hardest part is distinction->first filter has to separate fields out of the frame (ok that can be done )
than it must compare 3 fields at the same time!
[ i propose you download first zip file from my web (cause it looks like you didn't saw it ) you cannot speak "blindly".....blended fields are easily spoted.......they appear only on motion...THEN you may post this info on asynth forum and ask developers if anything could be done....(i can see that there are new filters there every day...hehe)
btw.i have no interest in such filter->manga means nothing to me!
but there are people that would benefit from this! ]

d>In short you cannot recover the lost data, all you can do is blend away the effects. This is just a fact.

_yes,but interpolation may help....


d>the only problem i could see is with native uk/european pal tv stock (not film) but then i bet you only get frame duplication and not blending (if at all).

_in standards conversions fields are manipulated,not frames!
you're mixing it!
cause tv works on field,NOT frame basis!
field duplication?

pal->ntsc has to introduce new fields(!)
(50hz->60hz)
ntsc->pal has to "throw away" fields(!)
(60hz->50hz)

either way interpolatiton is done in moving areas (in good converters)because if you simply "throw away" then motion is lost->NTSC captures 60 fields of tv information.throw anythin away and it's visible...particulary in motion scenes!!
if (ntsc->pal) has to throw something (and obviously it does!) away,then it MUST compensate for it by producing intermediate fields
that "cover" the motion
this way "judder" is reduced to minimum!
(manga's have maximum of judder)

pitty i didn't captured anything from football mondial (in Korea)
or recent olympics in us.....
i wonder what (hardware on-line) converters they used!

S>And is it possible to easily tell the difference between blended fields made by a poor NTSC-->PAL conversion and progressive PAL with a simple field shift? Both look interlaced (combed) on the monitor.

_because you look at them at frame basis on monitor!
use the script i mentioned,and in "phase-shift" there shouldn't be any ugly looking fields--->phase shift only messes with where fields end up,not with fields itself!



cheers

Ivo