(Resize and Deinterlace) or (Deinterlace and Resize)?

[Jaester]

Ok, what's better:

1 - Resize and then deinterlace
2 - Deinterlace and then resize

Any opinions?

Jaester

[Tuning]

De-interlacing after resize would be better.Then de-interlacing filter need only perform operation to reduced/Resized frame sizes.So less CPU load.

[Jaester]

But what about quality concerns? If deinterlacing only uses the information in two consecutive pixels it doesn't matter, but I believe that it uses info from consecutive adjacent pixels too, so resizing will reduce information used on deinterlacing. Plus, the resized frames will not be perfectly interlaced like the original with the lines splitting evenly.

Any thoughts?

Jaester

[Justinus]

I think that it's fine if you resize horizontally but not vertically before deinterlacing.

[mustardman]

You should always de-interlace BEFORE re-sizing. As you point out, trying to resize first screws up the interlacing badly, and can't be reversed.

Of course it depends what you want to do -
Horizontal resize only
Horizontal & vertical resize
Clipping/cropping (completely different from resize)

If you want to resize to somthing other than half original size, then AVIsynth could help...
For example :
De-interlace -> double frame rate
Resize every new frame (aka: used to be fields)
Re-interlace -> back to original frame rate

Why you would want to do this???

(except PAL <-> NTSC conversion, then there are pulldown issues as well!)

[reepa]

You don't need to deinterlace at all if you halven the x and the y resolutions. If you need to resize from PAL to NTSC for example, deinterlace first, then resize, like mustardman said. You can also separate the fields before resizing and weave them back together afterwards, and then deinterlace.

[theReal]

Any opinions?

This is not a matter of opinion, you must deinterlace before resizing, otherwise you turn an interlaced picture with two seperate fields into some combed mess that cannot be deinterlaced anymore (of course a deinterlace filter can still remove some of the combing, but you will always have a lot of artefacts left)

[Wilbert]

Of course you should separatefields, resize, weave in that case :)

[johnboy00]

And of course I confused "resize" with "crop" before voting :).

[vhelp]

Dealing w/ all sorts of sources files for several years now, I would
definately "opinionate" that quality'wise, deinterlacing first is key.

You see, if you deI last, you loose info, or will work w/ corrupt/disturbed
info. And also, when you resize, it has the "double-edge sword" effect too.

* PRO: in some situations, resizing can give the effects of "filtering or smoothing"
the source to some degree. This can be a good thing (first side of sword)

* CON: but, given the above PRO, by performing this step first, and deI later,
you will be destroying ALL of the pixel info that would be beneficial during
the deI first.. the temporalness or what-have-you.. (other side of sword)

Hope that made some sense.

On the other hand, although deI'ing does not "destroy" info, rather it results
in lesser info, due to the temporalness that is subtracted from, by the removal
of a field.
.
.
Assuming, we are talking about pure Interlace source :)

Just to note, like it or not, when dealing w/ pure interlace, (again, assuming)
we are talking about, for instance, DV cam footage. These are actualy half frames
or (720 x 240) sources. Those 720 x 240 are Interlaced into "one frame" hence
the new resoluiton of 720 x 480 (or, illusion thereof) ..still not a lot of resolution
when you really look at it from a "field" point of view. Then, you go out and perform
a deI, and loose that Interlaced x240 (or 2*240=480) frames, dropping a field,
and resizing to x480 of which is usually done during the filtering (inside the filter)
behind the scenes. Most user are abilvious to this (till now) :)
.
.
Building on this, (above) there are a number of way to "improve" upon this deI
step, to further the quality (that is) give the illusion of, more quality from the
loss of one field. It's tricky to do, but can be done. And, only to the trained
eye, can one disern the true nature of the final source (that it was deI'ed) :wink:
.
.
But, it takes time to build these kinds of experiences. And, the casual user will
not understand this.. nor be able to produce better quality from a deI'ed source.

-vhelp

[Jaester]

Vhelp,

Come on, enlighten us, please?!

What are the number of ways to improve the DeI process and get that high quality we all love and seek? And, yes, we are talking about DV cam footage. :)

Jaester

Originally posted by vhelp
Dealing w/ all sorts of sources files for several years now, I would
definately "opinionate" that quality'wise, deinterlacing first is key.

You see, if you deI last, you loose info, or will work w/ corrupt/disturbed
info. And also, when you resize, it has the "double-edge sword" effect too.

* PRO: in some situations, resizing can give the effects of "filtering or smoothing"
the source to some degree. This can be a good thing (first side of sword)

* CON: but, given the above PRO, by performing this step first, and deI later,
you will be destroying ALL of the pixel info that would be beneficial during
the deI first.. the temporalness or what-have-you.. (other side of sword)

Hope that made some sense.

On the other hand, although deI'ing does not "destroy" info, rather it results
in lesser info, due to the temporalness that is subtracted from, by the removal
of a field.
.
.
Assuming, we are talking about pure Interlace source :)

Just to note, like it or not, when dealing w/ pure interlace, (again, assuming)
we are talking about, for instance, DV cam footage. These are actualy half frames
or (720 x 240) sources. Those 720 x 240 are Interlaced into "one frame" hence
the new resoluiton of 720 x 480 (or, illusion thereof) ..still not a lot of resolution
when you really look at it from a "field" point of view. Then, you go out and perform
a deI, and loose that Interlaced x240 (or 2*240=480) frames, dropping a field,
and resizing to x480 of which is usually done during the filtering (inside the filter)
behind the scenes. Most user are abilvious to this (till now) :)
.
.
Building on this, (above) there are a number of way to "improve" upon this deI
step, to further the quality (that is) give the illusion of, more quality from the
loss of one field. It's tricky to do, but can be done. And, only to the trained
eye, can one disern the true nature of the final source (that it was deI'ed) :wink:
.
.
But, it takes time to build these kinds of experiences. And, the casual user will
not understand this.. nor be able to produce better quality from a deI'ed source.

-vhelp

[vhelp]

hi jaester,

funny how's you talk about enlighten, I had just finished w/ what I thought to
to be an "improve" result, using various filters in vdub, I saved them,
and closed down vdub, went into vdub and reloaded up my so called template.
To my surprise, the template was blank :eek:

I spent this whole evening tweaking it, and now it's all gone !!
And, I'm too burnt out and tired from all this tweaking around, and
for nothing :(

The last clip I was testing the above template took 13 minutes to encode
a 1 minute clip. You have to weight the benefits vs. time vs. quality
vs. what
you're willing to go far in.

As for me, I'm mostly debuggin for the best possible method of deI'ing
pure interlace sources for when I share small clips from my Video Scrapbook
collections :)
Yeah, these are various random shots of scenes that I happen to capture,
or short events etc. and I put them in divX and share. But, every so
often, I like to see how it would look when I encode them under TMPG
in 16:9 w/ 2.35:1 ratio etc. That last test clip, I thought came out
pretty good. No, I didn't say as good as Interlace. just better than
those source using those Deinterlace filters (imo)

When I'm in my template dubuggin stage, I start w/ a generic template
and work my way w/ other filtering steps, and mostly do a lot of slow
"trial n error" encodes. Very time consuming, but being that this is
a hobby of mine, time (or waist thereof) is not an issue :)

Maybe tomorrow, I'll try again at it - if I can remember some of the
steps I took and the filtering I used. Until then..

Good luck w/ your technique(s) :)
-vhelp

[Jaester]

vhelp,

Can you please give a brief overview of your methods/tools? I ask this because I have only recently become interested in tweaking my DeI and don't really have the time to keep testing stuff, my hobbies are other than this.

Maybe I'll brief you on my whole process and you can point me to problems/enhancements: capture on premiere, edit on premiere (with recoding turned off), export as DV, then deinterlace-crop-resize and enconde in VirtualDub.

I deinterlace with the Blend Fields together method. What I am not sure is if I should separate the filds before doing this or that is all done for me by the filter.

Usually I keep the original size but crop it sometimes I want to reduce it a little to get a smaller encoding size.

Jaester