Deinterlacing Frustration - Light Show with Strobe

[Bob Richardson]

Nearly 15 years ago I was involved in an outdoor light show production for a public event, and another one two years later.

Although I managed documenting the show on video back at the time, I have not been in possession of the master tapes until recently. Now, for reasons both of nostalgia and because several of the original team are interested in securing funding to do a new show, and because the master tapes were finally found (in a crawlspace -- not mine!), I'm re-editing these in FCP.

The original tapes were S-VHS and Beta SP, NTSC (interlaced, of course). I had a production house transfer these to ProRes 422(HQ) NTSC (still interlaced) -- the point was to get the best possible transfer without introducing any conversion artifacts. The transfer looks fine to me (considering the age and quality of the original recordings).

But, for various reasons (primarily because we are posting the video online), I want to convert the footage to progressive. (Please don't try to talk me out of this.) I'm running into a great many headaches with this.

Here's the root of the problem:

At many points during the light show, there is an effect done with a strobe light. The strobe is fast, and sometimes it only shows up in the Even or Odd field, but not both.

If I place my interlaced clip on a progressive timeline in FCP7, and apply no filters, FCP7 creates a progressive frame by merely throwing away one of the fields, and then doubling the lines in the remaining field.

Not only does this look terrible, resolution-wise, but it also removes many of the strobe effects altogether (because they were only visible in one field).

I thought I could solve this by running the whole production through Compressor first, because it has more advanced deinterlacing controls.

In Compressor, if I turn Frame Controls on, output fields set to "Progressive", I have three choices of deinterlacing methods. The first, "Line Averaging" actually just seems to do the same thing that Final Cut does -- throws away half the information, and loses many of the strobe effects. The other two, "Better - Motion Adaptive" and "Best - Motion Compensated" look much better in terms of resolution, but the strobe effects are still mostly gone -- I'm thinking that because they occur in only one field, the deinterlacing algorithm is dismissing them as noise.

So, back to Final Cut. I've tried any number of deinterlacing filters on the timeline, from FCP's own "deinterlace", to TMTS, Joe's Filters, Nattress, and ReVision. All have slightly different results some have a wide array of controls, but I keep winding up with results that look like half the resolution is gone, or you can't see the strobe.

All I really want is for all the scan lines in a complete interlaced frame (two fields, even and odd), to be rendered into one progressive frame.

I don't care if there are motion artifacts or the usual things that bother people about interlaced footage. In this case, the camera is locked down, the entire presentation is an abstract light/music show. Losing resolution or losing the strobes is not acceptable.

Any advice? If my description above isn't clear enough, let me know and I'll take some time and put together some sample clips which should make the story obvious.

All I really want to do is force FCP to use every available scan line when making a progressive frame, no other changes.

Late update: The program JES Deinterlacer is providing much better results than the other methods, but it can only output to a limited array of formats. I'm afraid that introducing a format conversion (to Apple Intermediate Codec in this case) is slightly degrading the video over staying in ProRes. And JES also seems to be playing gamma and cropping just a tad.

Thanks in advance!

[R Geoff Baker]

If the strobe only shows up when you have 60 fields per second ... how can they preserved if you drop the rate to 30 per second? Seems to me you'd have to create a 60p file, not a 30p one -- no? And therefore you must use a line doubling method, as the adjacent fields have unsuitably different material in them ...

Cheers,
GB

[Bob Richardson]

Yes, there are 60 fields per second, representing 480 visible scan lines. (2 240-line fields in one frame.)

Thus, a single 480-line 30FPS progressive frame can contain all of the scan lines from two fields with no loss of resolution.

The fields come from two slightly different moments in time, which can cause motion artifacts which are usually undesirable. But in this application what's more undesirable is losing any information from the fields.

[R Geoff Baker]

I can see how you hate to give up the resolution, but if the strobe effect is happening at faster than frame speed -- which is what you conclude when you say it is only visible at field speed -- I don't see how you can have it both ways: Merge two fields and save the resolution but lose the strobe effect (lose the temporal resolution) or line double a field and lose the vertical resolution but preserve the temporal resolution (strobe effect).

Cheers,
GB

[Bob Richardson]

I do appreciate the replies, but I am reasonably familiar with interlaced video and the limitations inherent in conversion to progressive. I do know that I can display 480 lines from two 60i fields as one 30p frame as long as I'm willing to accept some of the artifacts.

What I'm looking for is the specific technique to do this via Final Cut Pro or Compressor. As I mentioned, I already found a 3rd-party utility (JES Deinterlacer) that does a reasonable job, but it doesn't save to ProRes.

Trust me, it really is possible to convert interlaced to progressive and maintain 100% of the recorded resolution. I just need to know how to get the results I want with the tools I'm currently using.

[R Geoff Baker]

Sorry I can't help you -- and you don't seem to like what I'm trying to say ...

You can EITHER preserve vertical resolution or temporal resolution but not both. It's physics, not filter -- and a careful reading of your first post seems to confirm that you've seen that with the filters you've used, and your own statement that you know this ...

So I guess I don't understand what you want -- any filter that preserves vertical resolution does it by using the data from temporal resolution, any filter that preserves temporal resolution does it by restricting the data to that time/field. As you say 'results that look like half the resolution is gone, or you can't see the strobe' ...

Cheers,
GB

[Bob Richardson]

That's just it Geoff -- I don't *want* to use a filter. I want both fields in one frame, unfiltered. Final Cut won't do this unassisted, so I tried various options in a variety of filters to see if I could make it happen.

I don't know how many times I have to say I'm willing to accept motion/temporal artifacts.

That's fine.

I just want all my scan lines, and I want them in one progressive frame.

This isn't rocket science.

[R Geoff Baker]

My confusion then -- I understood that when the two fields were combined to one frame, the strobe effect was lost & you didn't like that. I think I'm understanding that you do want the effect of two fields blended to create one frame, no matter the consequences to the strobe effect ...

Which is what I understood the Natress filters to do -- he has four I think, but the 'blended' option does just that I think. Or Piero Fiorani's more complicated time remapping filter -- you can choose the method that blends the fields but set the remap function to none ...

Both Natress & Fiorani have always been prompt in responding to my questions or concerns when messaged directly. Not sure that this can be done _without_ using a filter -- algorithm, filter, instruction set ... what's the difference?

Cheers,
GB

[Bob Richardson]

Thanks, I will contact them and report back anything pertinent.

[William Hohauser]

Have you experimented with 480 60p or 480 30p timelines? These would treat the strobe fields as their own frames or blend the two together. These are not standard 4:3 formats but it might get the base video ready for transcoding to a web format. Unfortunately blending the fields will cause the strobe to lose half of it's intensity while the rest of the frame would remain the same, while the 60p will have lower resolution due to the fact that a field is 240 lines.

Never tried this, it's a thought experiment at this point.