Deinterlacing Question

[the_fusionist]

When you shoot 1280 lines at 50i fields per second would that not play at 640 interlacing lines which would amount to 1280 lines at 50i? And when you deinterlace that would that not be 1280p at 25 frames every .5 seconds? That means one would have to slow it down by 50% to equal 24p of 1280 lines.

1280 at 50i = 640 lines at 50i (only 640 lines are shown in every field)
1280 at 24p = 1280 lines at 24p every .5 seconds

*To get 24 frames a second would mean one would have to slow down the video by 50%.

Let me know if I have this right.

 

[the_fusionist]

Somebody? Anybody?

 

[reflex]

You're just trying to convert 50i to 24p?

 

[Proteus]

First of all there is no 1280 lines in any format. In HD there is 720 horizontal lines (1280x720) or 1080 (1920x1080) and the Sony cameras shoot only at 1080 horizontal lines at 50/60i. That is 50 for European, 60 for US models.

(There is also another way to refer to "lines" as when you need to describe the total definition of a camera in "TV lines" including the lens like when you shoot test charts, but that's another thing).

If you have a camera that shoot at 1280x720 then this is already progressive either at 24p, or 30p -no need to deinterlace.

Using the European Fx1 or a Z1 @ 50i you can convert the 50 fields to 50 frames (scale 2x vertically) and slow down to 24 or whatever only if you want slow motion.

If you just want to convert to 24p you just have to omit the intermediate fields and convert the rest to frames to have 25p frames (this is the simplest way to deinterlace) but you will have much better results if you process it through a motion compensated adaptive deinterlacer instead. Then you can slow down by 4% to have 24p frames per second.

Now 1080 @ 50i means 540 horizontal lines (vertical resolution) in each field 50 times/second with a horizontal resolution of 1920 pixels (although the native horizontal res of HDV cameras is 1440). In 50/60i, moving objects have definitely only 540 lines of vertical resolution, but still or very slow moving objects have a perceptual resolution of 1080 lines.

 

[the_fusionist]

Sorry, I had the numbers wrong. I was thinking 1080i but writing 1280.

If you convert 50 fields to frames would that not add up to 25frames at full resolution? I read about people complaining but it should be the same as a progressive scan at 25 frames, then one could slow it down 4%.

I'm also thinking that after you deinterlace you end up with 25 frames at .5 seconds. If it took 1 second to show 50 fields then combining those fields would reduce time by 1/2. Unless I am thinking about this wrong?

Why would one need to scale 2x? Would that not bring up negative space?

I'll search the internet, but thank you for your reply.

 

[Proteus]

Fields, frames and deinterlacing

Fields, frames and deinterlacing

To answer your questions in full, it's necessary to explain in an organized way all the information that is needed:

The image is captured at regular time intervals say at t1, t2, t3, t4 and so on. In each time interval the image coming from the CCD is captured as a field not a frame. This means that only half the horizontal lines are captured (540 lines instead of 1080) by just omitting the intermediate horizontal lines. So the first field will contain the 540 odd lines 1,3,5,7......1077,1079 while the second field will contain the 540 even lines 2,4,6,8......1078,1080 of a frame.

Therefore in the 50i interlaced format we have:
t1dd lines field = first half of first frame.
t2:even lines field = second half of first frame, first frame completed
t3dd lines field = first half of second frame.
t4:even lines field = second half of second frame, second frame completed
and so on up to t50. Each time interval = 1/50 of a second.

In the 25p progressive format we have:
t1: all lines = first frame completed
------- the image from first frame is sustained-----
t2: all lines = second frame completed
------- the image from second frame is sustained-----
and so on up to t25. Each time interval = 1/25 of a second (double the duration of 50i interlaced)

Now how would you convert the above 50i format to much the 25p format?
If you shoot a subject with no motion, then the interlaced will not have any difference from the progressive because the even and odd fields will alternatively project in a CRT TV very fast (every 50th of a second) and your eyes will hold the information from each field long enough (because of "aftervision") and will nicely combine them to see the full frame (this is done digitally in modern sets). So in such a case you don't have to convert anything at all.

The problem is when there is motion (which is the nature of video after all) because a moving subject's information has been splitted in two different time intervals AND in two different places in the screen(!) and because of the time passed, each half will probably be from a different snapshot too! (e.g. subject rotated, it's parts moved or altered, lighted differently etc). Because of these differences, it's *impossible* to achieve a perfect interlaced to progressive conversion, so you have to compromise.

As I said before, each field differs from a full frame because it has half the horizontal lines which means half the vertical resolution, so a full frame's res is 1920x1080 and a field's is 1920x540. So the simplest way to convert the above 50i to 25p is to omit the intermediate fields and scale* 2x vertically the remaining, to have:

t1: odd lines field = first half of first frame.->2x vertical scale->a compromised first frame completed
t2:------- the image from first frame (t1) is repeated-----
t3: odd lines field = first half of second frame.->2x vertical scale->a compromised second frame completed
t4:------- the image from second frame (t3) is repeated-----
and so on up to t50. Each time interval = 1/50 of a second.

OR:

t1: odd lines field = first half of first frame.->2x vertical scale->a compromised first frame completed
------- the image from first frame (t1) is sustained-----
t2: odd lines field = first half of second frame.->2x vertical scale->a compromised second frame completed
------- the image from first frame (t2) is sustained-----
and so on up to t25. Each time interval = 1/25 of a second

To achieve the 24fps as in film, you just have to slow down the video and audio by 4%. You don't have to correct the pitch of the audio IMO because the difference cannot be heard without a reference -it's even less than a musical semitone (each next lower semitone differs by 5.6%).


Instead of just scaling each field, a more sophisticated method is to find the splitted parts of a moving object and combine as much of the information as you can to replace the resized pixels with information from both fields. Also, to recognize the non moving objects and avoid resize them at all -just to combine the information that belongs to these still objects from the two fields. This is the job of a good (and sometimes very expensive) motion compensated adaptive etc deinterlacer. Unfortunately the best deinterlacers are found in hardware devices where they are protected better with patents and from copying.

*the scaling is done by interpolating the neighboring pixels (combining information) so it won't bring up "negative space" actually, but you'll have an image with well hidden "redundant" information instead and of course less information than the actual frame -something more like a sharpened image from a blurred one.

I hope you understand it now.

 

[Proteus]

hmmmm..... maybe you don't!

 

[the_fusionist]

oh no, I got it. You did a good job of explaining it. I was thinking that 50i would be the same as 25p, but the fields are disjointed. So in essence 50i contains more information but less in each field while 24p has less information and a whole frame of it. The lower frames creates the illusion of film. I think I will buy a dvx100b and forget the fx1 untill they decide to go progressive in the same package.

 

[the_fusionist]

I don't know if you got that, but it could be just semantics.

 

[Proteus]

So in essence 50i contains more information but less in each field while 24p has less information and a whole frame of it

Simply, 50i offers double the information about the motion of a *moving* object but half the information about how this moving object looks, while 24p offers the opposite.
Static objeccts have the same information (zero on motion and max on definition ) in both formats. Note that the total infomation between 50i and 25p is the same.

The lower frames creates the illusion of film.

What I meant is that when the second field has been projected, this means that the total information of a frame has been projected, because a frame=2 fields. Now if the upper fields are omitted and you make the lower fields frames with scaling then these frames will project in rates similar to film.

I think I will buy a dvx100b and forget the fx1 untill they decide to go progressive in the same package.

DVX100b is not HD, and not 16:9 either. So I don't think they are comparable.

Personally I like progressive but only 60p HD! The best from both worlds. But we are far from this (from a good implementaion for capture, storage and playback)
...And I wouldn't say no for 100p! -just dreaming... the "reality look"!

.

 

[the_fusionist]

What I meant is that when the second field has been projected, this means that the total information of a frame has been projected, because a frame=2 fields. Now if the upper fields are omitted and you make the lower fields frames with scaling then these frames will project in rates similar to film.


DVX100b is not HD, and not 16:9 either. So I don't think they are comparable.

Personally I like progressive but only 60p HD! The best from both worlds. But we are far from this (from a good implementaion for capture, storage and playback)
...And I wouldn't say no for 100p! -just dreaming... the "reality look"!

.

what I meant was that the lower frame rate 24p creates the illusion of film. Thanks for the advice. I will debate what to do. What deinterlacer do you think I should use? I use Vegas as an editor.

 

[Proteus]

What deinterlacer do you think I should use? I use Vegas as an editor.

I haven't done a recent search for a specialized app so I recommend you to search and then do exhaustive *tests* for each software app that promises good results. For example, slow stable pans in the same environment and conditions, or shooting a fan that rotates slowly.

These tests are especially needed if you want to find a deinterlacer for 60i to 24. A good deinterlacer should work on field level and select 24 fields from 60i and then process them to make them frames. A bad one might deinterlace first to 30p and then select 24 frames. In the first case the resulted motion will be 2x smoother. If you want the best results, the deinterlacer should also be adaptive motion compensated.

One of the best but difficult for beginners solutions is Avisyth (free). This is a generic post-processing engine (without user interface -but you don't need one!) with the ultimate flexibility because you can make simple scripts (in notepad, ending in .avs) with a sequence of processing actions including adjustments, to process the video the way *you* want in one pass. You can do almost anything (including cropping resizing etc) with a single script but it requires that you have a solid understanding of the video format. Also you have to invest several hours to learn the basics for Avisynth and find the (free) plugins that will work for you but the forum is very good. And it's worth the effort.

Another free very good solution is VirtualDub Mod Mpeg2 (you have to collect plugins),which can frame-serve too in some cases. The Mpeg2 (unofficial) version can accept the raw HDV format after you do a quick conversion with the HDTV2MPEG2 free app.

A shareware app easy to use (but I haven't tested the quality) is DVFilm maker.

 

[the_fusionist]

Hey proteus, I have decided that I will spring for a hvx200. I can use it with vegas via Raylight. All this technical stuff is not for me. I understand it but I will put up the money for the HVX rather than spend time fiddling with anything complicated. I prefer to spend that time writing and shooting. Thanks for you help.

 

[Proteus]

You welcome! You will sure avoid deinterlacing with HVX200! It will probably cost you a lot more, but hey, those who can buy time... live more! And if your 1st priority is to write and shoot as you say, then you'll succeed! Good luck with your choice!

 

[Heath McKnight]

Final Cut Pro 5 users shooting in 50i, try this (thanks to Graeme Nattress):

1. Either shoot in 50i with CineFrame 25 or no CF25 (you'll need to de-interlace, though).

2. Edit in FCP 5 on an HDV 1080i50 timeline.

3. Copy and paste the project into a new sequence and select a non-HDV hi def codec. Render and export as a QuickTime movie, not Conversion.

4. Convert to 23.98 in CinemaTools.

5. Re-import into FCP and into a 24p timeline. Render and output.

Heath

 

[mvb]

Final Cut Pro 5 users shooting in 50i, try this (thanks to Graeme Nattress):

1. Either shoot in 50i with CineFrame 25 or no CF25 (you'll need to de-interlace, though).

2. Edit in FCP 5 on an HDV 1080i50 timeline.

3. Copy and paste the project into a new sequence and select a non-HDV hi def codec. Render and export as a QuickTime movie, not Conversion.

4. Convert to 23.98 in CinemaTools.

5. Re-import into FCP and into a 24p timeline. Render and output.

Heath

If you pre-convert your clips to 25P by dropping them into DVFilm Maker, you won't have to go through the deinterlacing and slowdown process every time you render a new cut. You edit and view at 25P. 25P has the same look as 24P.

Shooting Cineframe 25 greatly reduces vertical resolution vs. shooting 50i and converting to 25P. It's especially noticable on static, wide shots with a lot of fine detail.

 

[futureman]

I was just wondering... theoretically if you set your frame rate to 1/25th of a second on an fx1e recording in 50i, would it not already be deinterlaced, since both fields of the interlaced frame would take up the same time?

just a thought

 

[mvb]

I was just wondering... theoretically if you set your frame rate to 1/25th of a second on an fx1e recording in 50i, would it not already be deinterlaced, since both fields of the interlaced frame would take up the same time?

just a thought

On most cameras if the shutter time is more than the field interval it sums successive fields together to get the longer exposure, instantly knocking down your resolution.

 

[Heath McKnight]

If you pre-convert your clips to 25P by dropping them into DVFilm Maker, you won't have to go through the deinterlacing and slowdown process every time you render a new cut. You edit and view at 25P. 25P has the same look as 24P.

Shooting Cineframe 25 greatly reduces vertical resolution vs. shooting 50i and converting to 25P. It's especially noticable on static, wide shots with a lot of fine detail.

I would say it's not too bad, plus Marcus at DVFilm has stated you only lose around 20% resolution. I'd suggest doing tests both ways, 50i and 50i with CineFrame 25 and view both on an HDTV.

heath