Deinterlacing

Kin Kwan · April 3rd, 2005, 06:26 PM

It was shot on an Optura 20. :D

James Lilly · April 4th, 2005, 02:04 PM

What kind of resolution loss do you experience using these methods? Say taking it from 60i to 24p?

Jonathon Wilson · April 4th, 2005, 03:04 PM

It really depends.... the framerate doesn't really matter much -- think only about vertical lines:

When you take a standard NTSC source, you've got 29.97 frames per second, which gets chopped up into 59.94 fields per second. Each field has roughly 240 lines of resolution. (NTSC is actually 486, I believe).

When you deinterlace in avisynth, the 'Bobbers' use interpolation to recreate the missing lines in the fields. This results in 59.94 full frames per second with 480 lines each. However, the 480 lines were created by interpolating, so you do lose some vertical resolution -- depending on the smarts of the deinterlacer. In avisynth, there are a wide variety of different filters which can do this interpolation in a whole host of ways -- from just doubling the lines (simplest -- worst-looking) to some serious math, using both temporal and spatial information to recreate information that varies depending on whether the section of the frame is moving or not, etc. There are some wicked-smart deinterlacers for avisynth.

The subsequent conversion from these 59.94 progressive frames down to 24 progressive frames is a separate step which also has a wide variety of possible ways to do it, from the simplest -- just chop out enough frames to end up with 24 (usually 23.976) -- or much more complicated. The best results come from using 'temporal interpolators' which can actually recreate additional interpolated frames in between the ones which were actually shot. These are great for super-smooth slow-motion as well as recreating realistic motion blur. My favorite down-convert to 24p interpolates 10x the number frames in the original 60-frame progressive source (after deinterlacing). This give you a 600 fps(!) progressive source. Then carefully blending 25 frames together gives a motion-blurred frame which is very close to a 1/48th shutter speed. I keep 24 of these per second resulting in a glorious-looking 24p from NTSC input.

The best part is -- all of this is controllable... if I want to smear together a full 1/24th second exposure of blurred frames, I just blend 50 frames instead of 25. Or -- If I want to vary the blending such the the beginnings and endings of the blur are more transparent than the middle -- I can do that too.

Avisynth! But man... does it take some TIME to render! And -- it takes some learning to figure out how to make it do what you want. It helps if you have some programming in your background.

If people are interested, I'd be happy to throw together a bit of a tutorial on the steps I use on my web site. (I'm not saying I'm an expert, but my ideas might give you starting points for improvements, etc.) Maybe Kin would be interested in contributing -- a joint source of Avisynth goodness!

Riley Harmon · April 4th, 2005, 03:11 PM

Jonathan, I would love for you to write up a tutorial, it would be great and very appreciated.

James Lilly · April 4th, 2005, 05:00 PM

A tutorial would be most appreciated. I really found your explanation informative. Unfortuantely, I have a couple more questions then:)

How does 24p arrived at this way compare to cameras like the XL2 and the Panasonic's native 24p? In both image quality and in resolution?

Why do sony specs indicate 530 lines of resolution if it is really 480?

Do you know how Vegas Videos 24p convertor compares?

Thank you once again for you detailed and informative response, I'm learning quite a bit today.

Joshua Provost · April 4th, 2005, 05:11 PM

James,

Difference between this and the XL2/DVX100A? I think it helps to clarify a comment earlier in this thread. NTSC does not have 29.97 frames per second that are broken up into 59.96 fields per second. It's the other way around. NTSC cameras is capturing images at 60 (come on, let's round) separate moments each second. A single frame is a combination of two moments, 1/60 second apart, thus the interlacing artifacts.

XL2/DVX100A are capturing images at 24 full frames per second.

Bottom line, you can't beat the quality of a camera that is natively doing 24p. It will be clearer and crisper than any deinterlacing and/or interpolation. You can get close, but you are essentially trying to create data that is not there.

If you see a spec of ~530 that is referring to the horizontal resolution. NTSC DV is 720x480. So it's 530 out of a possible 720. Vertical resolution, in reality, may be somewhere in the 300's.

Josh

Jonathon Wilson · April 4th, 2005, 05:41 PM

I'll write the tut as soon as I get a free evening -- It will likely be this upcoming weekend, as of now. Pretty busy week!

A 'real' 24p capture will likely always be superior because you've got 24 progressive frames of full resolution to start with. Any time you de-interlace, there has to be some interpolation which tries to figure out the missing 'in-between' lines, so a full-progressive capture will always have an edge here. How much edge... it varies on your source. The most difficult things will be sources which have a high need for vertical resolution -- very fine horizontal lines, etc. These won't do well in the interped version, where the progressive cap should see them.

The sony specs you're talking about are probably 'TV Lines'. TV Lines are a measure of horizontal resolution as a factor of aspect ratio. They're tricky, but in a nutshell, it's the number of 'identifiable details' in a single line divided by the aspect ratio:

Let's say you have something with 600 TVL of resolution -- shot at a 4:3 aspect ratio, you have a theoretical ability to resolve 600 * 4/3 or 800 horizontal pixels per line. This also gets influenced by other things, but generally, TVL are seen as a measure of horizontal resolution (columns) which are lower than the number of pixels. The higher the aspect ratio, the bigger the difference. Unfortunately, it also takes into account things like bandwidth of transmission, so you'll often see widely differing numbers.

So in answer to your question -- the 480 is the vertical resolution (in lines from top to bottom) -- the 530 is likely TV Lines which in 16:9 would be around 900 pixels, but it's not a straight conversion... search for "TV Lines" there's quite a bit of information.

http://jkor.com/peter/tvlines.html
http://members.aol.com/ajaynejr/vidcolor.htm

Btw: 'broadcast' cameras (as opposed to the 'Prosumer' cams most people here use) can turn in TVL resolutions in the 900s (!). There really is a difference between these $5000 cams and the $50,000 cams.

I'm afraid I have no experience with Vegas... other than to know that lot of people like it :) It's hard for me to believe that it 'outpowers' avisynth, but its probably much much easier to use -- and may produce perfectly fine results.

Jonathon Wilson · April 4th, 2005, 05:43 PM

Josh is exactly right on the fields vs. frames thing -- I was thinking in terms of my captures which grab 30 frames per second -- each of which has two fields-worth of data. The NTSC signal is fields. Sorry if I misled...

James Lilly · April 4th, 2005, 05:43 PM

Thank you Josh, I have always wondered about that. And this is, to be honest, one of the reasons I have just avoided the whole 24p debate with my cameras and my crew.

But now I have another question, how many lines of resolution are the 24p cameras putting out? Per frame? And what happens to the 24 p when burned onto a DVD and shown on a TV or a projector? My projector shows it at either 480i or 480p, what exactly does that mean?

Specifically, is it refering to 480 lines of resolution rather than frames per second?

Jonathon Wilson · April 4th, 2005, 05:49 PM

By the way, Josh, are you interested in sending your favorite deinterlacing methods in Avisynth? I'll be happy to include anything you send into this little web article this weekend...

Joshua Provost · April 4th, 2005, 06:31 PM

Quote:

Thank you Josh, I have always wondered about that. And this is, to be honest, one of the reasons I have just avoided the whole 24p debate with my cameras and my crew.

Since you don't have a 24p camera, you don't have to bring it up with cast and crew. It will all be what you do in post, so you can keep it quiet. ;)

Quote:

But now I have another question, how many lines of resolution are the 24p cameras putting out? Per frame?

The resolution recorded to the tape is the same on 60i or 24p cameras: 720x480. The real resolution? Check out these resolution test charts.

Panasonic GS400: http://www.pana3ccduser.com/images/u...ed_eia1956.jpg
Canon XL2: http://www.dvxuser.com/articles/shoo...%20EIA1956.JPG
Panasonic DVX100A: http://www.dvxuser.com/articles/shoo...p-thin-4x3.jpg

Quote:

And what happens to the 24 p when burned onto a DVD and shown on a TV or a projector? My projector shows it at either 480i or 480p, what exactly does that mean?

This refers to lines, vertical lines. NTSC video (720x480) is the same as 480i.

When you burn a 24p DVD, the DVD player adds 2:3 pulldown, a special sequence of fields, to generate a 60i/480i signal.

The advantage of a 24p DVD is it has fewer frames to encode into the same bitrate, so better compression quality as compared with a 60i DVD.

All this being said, I use DVFilm Maker, and I am very pleased with it. It samples every 2.5 fields (implies some field blending is done) to try to work with the exact moments in time that you would have captured were you shooting 24p or film.

It's not terribly "intelligent" as deinterlacers go. It detects interlace artifacts (using configurable thresholds) and kills them using basic field doubling. So, it loses resolution, but only in the areas where there is motion, where your eye won't perceive the resolution loss, anyway. And it can smooth the jagged horizontal lines that may result.

It's very simple and very fast, and I like that it is a stand-alone app. It is well worth the money.

Using my GS400, DVFilm Maker, a little color correction, and a lot of work on set, I have won a cinematography award recently, and have other work on a local short film cometition TV show. I've gotten great feedback and professionals asking me what I shoot with and how I got certain looks. You can do a lot with very little.

Also, I think people get caught up in comparing still images when discussing deinterlacing. The best still does not necessarily make the best image in motion. In fact, overly crisp stills resulting from 60i-24p conversion can cause noticable strobing. Blur/resolution loss can be your friend in certain siutations.

Also, any deinterlacer that interpolates can cause strange effects when dealing with detailed patterns. The combination of DV low resolution with interpolation can be a bad thing.

I've done five or six projects now with 60i to 24p, so I hope that is some helpful information from the field.

Thanks,
Josh

James Lilly · April 5th, 2005, 10:52 AM

Thank you very much for your detailed and techincal responses, I found them interesting and informative.

But they make me wonder even more things about the damn formats. I wish we could just pick the best one and gear up technology wise for it.

Kin Kwan · April 6th, 2005, 08:14 AM

Whoa, that's a good load of info. Yea, this interlace format does cause a lot of headache. I've been searching for a good method to deinterlace my videos ever since I started making movies with my DV cam, but I don't think I'm any where close!

Jonathon, I don't know a lot about programming, but I'd love to contribute to the tutorial. Doom9 has an awesome Avisynth forum with a LOT of experts over there. I'll send you my script for the method I used for deinterlacing. Should I send it to your email?

Jonathon Wilson · April 6th, 2005, 09:00 AM

Sure thing -- whatever you send I'll include... Anyone else have good avisynth deinterlacing scripts, feel free to email them to me. I'm hoping to put something up this coming weekend.

Joshua Provost · April 6th, 2005, 12:30 PM

James,

There are standards, and unless you get an XL2/DVX100A, you will most likely be shooting NTSC 60i/480i. In the end, the TV sets will use the same to present the image. What you do in between to give a film look is up to you.

I've put together DVDs with 60i and 24p versions of the same film, and there is a big difference. Until you see the two right after each other, it can be tough to understand, but it's big. Of course, not as big a film effect as shallow depth of field (see the Alternative Imaging forum).

Josh

April 4th, 2005, 03:04 PM	#18
Jonathon Wilson Regular Crew Join Date: Feb 2004 Location: Austin, TX Posts: 182	It really depends.... the framerate doesn't really matter much -- think only about vertical lines: When you take a standard NTSC source, you've got 29.97 frames per second, which gets chopped up into 59.94 fields per second. Each field has roughly 240 lines of resolution. (NTSC is actually 486, I believe). When you deinterlace in avisynth, the 'Bobbers' use interpolation to recreate the missing lines in the fields. This results in 59.94 full frames per second with 480 lines each. However, the 480 lines were created by interpolating, so you do lose some vertical resolution -- depending on the smarts of the deinterlacer. In avisynth, there are a wide variety of different filters which can do this interpolation in a whole host of ways -- from just doubling the lines (simplest -- worst-looking) to some serious math, using both temporal and spatial information to recreate information that varies depending on whether the section of the frame is moving or not, etc. There are some wicked-smart deinterlacers for avisynth. The subsequent conversion from these 59.94 progressive frames down to 24 progressive frames is a separate step which also has a wide variety of possible ways to do it, from the simplest -- just chop out enough frames to end up with 24 (usually 23.976) -- or much more complicated. The best results come from using 'temporal interpolators' which can actually recreate additional interpolated frames in between the ones which were actually shot. These are great for super-smooth slow-motion as well as recreating realistic motion blur. My favorite down-convert to 24p interpolates 10x the number frames in the original 60-frame progressive source (after deinterlacing). This give you a 600 fps(!) progressive source. Then carefully blending 25 frames together gives a motion-blurred frame which is very close to a 1/48th shutter speed. I keep 24 of these per second resulting in a glorious-looking 24p from NTSC input. The best part is -- all of this is controllable... if I want to smear together a full 1/24th second exposure of blurred frames, I just blend 50 frames instead of 25. Or -- If I want to vary the blending such the the beginnings and endings of the blur are more transparent than the middle -- I can do that too. Avisynth! But man... does it take some TIME to render! And -- it takes some learning to figure out how to make it do what you want. It helps if you have some programming in your background. If people are interested, I'd be happy to throw together a bit of a tutorial on the steps I use on my web site. (I'm not saying I'm an expert, but my ideas might give you starting points for improvements, etc.) Maybe Kin would be interested in contributing -- a joint source of Avisynth goodness! __________________ http://www.aqua-web.com

April 4th, 2005, 05:41 PM	#22
Jonathon Wilson Regular Crew Join Date: Feb 2004 Location: Austin, TX Posts: 182	I'll write the tut as soon as I get a free evening -- It will likely be this upcoming weekend, as of now. Pretty busy week! A 'real' 24p capture will likely always be superior because you've got 24 progressive frames of full resolution to start with. Any time you de-interlace, there has to be some interpolation which tries to figure out the missing 'in-between' lines, so a full-progressive capture will always have an edge here. How much edge... it varies on your source. The most difficult things will be sources which have a high need for vertical resolution -- very fine horizontal lines, etc. These won't do well in the interped version, where the progressive cap should see them. The sony specs you're talking about are probably 'TV Lines'. TV Lines are a measure of horizontal resolution as a factor of aspect ratio. They're tricky, but in a nutshell, it's the number of 'identifiable details' in a single line divided by the aspect ratio: Let's say you have something with 600 TVL of resolution -- shot at a 4:3 aspect ratio, you have a theoretical ability to resolve 600 * 4/3 or 800 horizontal pixels per line. This also gets influenced by other things, but generally, TVL are seen as a measure of horizontal resolution (columns) which are lower than the number of pixels. The higher the aspect ratio, the bigger the difference. Unfortunately, it also takes into account things like bandwidth of transmission, so you'll often see widely differing numbers. So in answer to your question -- the 480 is the vertical resolution (in lines from top to bottom) -- the 530 is likely TV Lines which in 16:9 would be around 900 pixels, but it's not a straight conversion... search for "TV Lines" there's quite a bit of information. http://jkor.com/peter/tvlines.html http://members.aol.com/ajaynejr/vidcolor.htm Btw: 'broadcast' cameras (as opposed to the 'Prosumer' cams most people here use) can turn in TVL resolutions in the 900s (!). There really is a difference between these $5000 cams and the $50,000 cams. I'm afraid I have no experience with Vegas... other than to know that lot of people like it :) It's hard for me to believe that it 'outpowers' avisynth, but its probably much much easier to use -- and may produce perfectly fine results. __________________ http://www.aqua-web.com

April 4th, 2005, 05:43 PM	#23
Jonathon Wilson Regular Crew Join Date: Feb 2004 Location: Austin, TX Posts: 182	Josh is exactly right on the fields vs. frames thing -- I was thinking in terms of my captures which grab 30 frames per second -- each of which has two fields-worth of data. The NTSC signal is fields. Sorry if I misled... __________________ http://www.aqua-web.com

April 4th, 2005, 05:49 PM	#25
Jonathon Wilson Regular Crew Join Date: Feb 2004 Location: Austin, TX Posts: 182	By the way, Josh, are you interested in sending your favorite deinterlacing methods in Avisynth? I'll be happy to include anything you send into this little web article this weekend... __________________ http://www.aqua-web.com

April 6th, 2005, 09:00 AM	#29
Jonathon Wilson Regular Crew Join Date: Feb 2004 Location: Austin, TX Posts: 182	Sure thing -- whatever you send I'll include... Anyone else have good avisynth deinterlacing scripts, feel free to email them to me. I'm hoping to put something up this coming weekend. __________________ http://www.aqua-web.com

April 3rd, 2005, 06:26 PM	#16
Kin Kwan Regular Crew Join Date: Jul 2004 Location: Plano, TX Posts: 120	It was shot on an Optura 20. :D

April 4th, 2005, 02:04 PM	#17
James Lilly Regular Crew Join Date: Jun 2004 Posts: 66	What kind of resolution loss do you experience using these methods? Say taking it from 60i to 24p?

April 4th, 2005, 03:11 PM	#19
Riley Harmon Major Player Join Date: Jul 2003 Location: Oklahoma Posts: 424	Jonathan, I would love for you to write up a tutorial, it would be great and very appreciated.

April 4th, 2005, 05:00 PM	#20
James Lilly Regular Crew Join Date: Jun 2004 Posts: 66	A tutorial would be most appreciated. I really found your explanation informative. Unfortuantely, I have a couple more questions then:) How does 24p arrived at this way compare to cameras like the XL2 and the Panasonic's native 24p? In both image quality and in resolution? Why do sony specs indicate 530 lines of resolution if it is really 480? Do you know how Vegas Videos 24p convertor compares? Thank you once again for you detailed and informative response, I'm learning quite a bit today.

April 4th, 2005, 05:11 PM	#21
Joshua Provost Major Player Join Date: Sep 2004 Location: Phoenix, AZ Posts: 493	James, Difference between this and the XL2/DVX100A? I think it helps to clarify a comment earlier in this thread. NTSC does not have 29.97 frames per second that are broken up into 59.96 fields per second. It's the other way around. NTSC cameras is capturing images at 60 (come on, let's round) separate moments each second. A single frame is a combination of two moments, 1/60 second apart, thus the interlacing artifacts. XL2/DVX100A are capturing images at 24 full frames per second. Bottom line, you can't beat the quality of a camera that is natively doing 24p. It will be clearer and crisper than any deinterlacing and/or interpolation. You can get close, but you are essentially trying to create data that is not there. If you see a spec of ~530 that is referring to the horizontal resolution. NTSC DV is 720x480. So it's 530 out of a possible 720. Vertical resolution, in reality, may be somewhere in the 300's. Josh

April 5th, 2005, 10:52 AM	#27
James Lilly Regular Crew Join Date: Jun 2004 Posts: 66	Thank you very much for your detailed and techincal responses, I found them interesting and informative. But they make me wonder even more things about the damn formats. I wish we could just pick the best one and gear up technology wise for it.

April 6th, 2005, 08:14 AM	#28
Kin Kwan Regular Crew Join Date: Jul 2004 Location: Plano, TX Posts: 120	Whoa, that's a good load of info. Yea, this interlace format does cause a lot of headache. I've been searching for a good method to deinterlace my videos ever since I started making movies with my DV cam, but I don't think I'm any where close! Jonathon, I don't know a lot about programming, but I'd love to contribute to the tutorial. Doom9 has an awesome Avisynth forum with a LOT of experts over there. I'll send you my script for the method I used for deinterlacing. Should I send it to your email?

April 6th, 2005, 12:30 PM	#30
Joshua Provost Major Player Join Date: Sep 2004 Location: Phoenix, AZ Posts: 493	James, There are standards, and unless you get an XL2/DVX100A, you will most likely be shooting NTSC 60i/480i. In the end, the TV sets will use the same to present the image. What you do in between to give a film look is up to you. I've put together DVDs with 60i and 24p versions of the same film, and there is a big difference. Until you see the two right after each other, it can be tough to understand, but it's big. Of course, not as big a film effect as shallow depth of field (see the Alternative Imaging forum). Josh