Capture/record in progressive or deinterlace? [Archive]

View Full Version : Capture/record in progressive or deinterlace?

Jumbie

8th October 2006, 16:08

Hi guys (and gals) just got my hands on a new camera that is capable of recording in progressive mode and was wondering if you could help me out with this question.

To cut straight to the chase, the g/f and I are into making "home movies". These are only viewed on a computer screen due to the fact that I can keep them encrypted when they are not in use. Don't want to have DVDs lying around for someone to find. :eek: Therefore, I don't need interlaced footage that will, from my reading, look better on a TV screen and I've always deinterlaced in the past.

The camera will be on a tripod so stabilization is not an issue. I also will be capturing/recording directly to a laptop hard drive via WinDV through firewire.

Here is the description of the progressive mode from the manual (camera is a Sony DCR-PC330):

This feature is useful for movies you intend to import to your computer to produce still images.
You can record pictures with less blurring than recording in normal mode, producing pictures suitable for analyzing high-speed action such as sports. To enjoy a movie, it is better to record the picture in normal mode, since the image is recorded only evern 1/30 of a second in the progressive rocording mode.

Note on the progressive recording mode
In a normal TV broadcast, the screen is divided into 2 finer fields and these are displayed in turn, every 1/60 of a second. This, the actual picture displayed in an instant covers only half of the apparent picture area. In progressive recording, the picture is fully displayed with all the pixels. A picture recorded in this mode appears clearer, but a moving subject may appear awkward.

So, the last line of the above quote about the moviing subject appearing a little awkward has me a little concerned (we can just test it I guess) but I'm also curious as to whether recording in progressive will affect the compressibility of the video when I encode it to x264.

Any advice on what may be best? Should I stick to deinterlacing normal footage? It's just that I'm not aware of the advantages and drawbacks of progressive vs. normal interlaced footage.

Thanks in advance.

Blue_MiSfit

9th October 2006, 00:28

what kind of camera is it?

The Panasonic DVX100 is very popular for this and records progressive in 30p or 24p, though it wraps these in 60i through pulldown. Removing it is a little tricky but not too bad.

My thoughts on interlaced vs progressive are that if you need to record high motion content (sports, cars, action sequences etc) then 60i is the ONLY way to go, unless you can shoot 60p :D

30p looks really shuttery and bad for high motion stuff.

Of course it depends on your target. If you are going for a DVD on TV, then interlaced all the way!! If you are targeting a computer then you have to deinterlace sometime.. :)

~MiSfit

Jumbie

9th October 2006, 01:05

what kind of camera is it?

I mentioned it in my first post. ;)

It's a Sony DCR-PC330. Not the greatest but better than what I was using before. Can't afford anything better a.t.m. Maybe when I finish medical school in a couple years. Well make that more like 7 years by the time I finish residency and can atually earn something...

The Panasonic DVX100 is very popular for this and records progressive in 30p or 24p

Well, from what the manual says, it seems to be 30p. I've included the blurb on progressive recording above.

though it wraps these in 60i through pulldown.

Don't believe that's the case with my camera.

My thoughts on interlaced vs progressive are that if you need to record high motion content (sports, cars, action sequences etc) then 60i is the ONLY way to go

Well, like I said before I was deinterlacing and it seemed fine for what I was recording. I wasn't doing anything fancy. In fact, I think I was just discarding one of the fields.

unless you can shoot 60p :D

Sadly, no.

Of course it depends on your target. If you are going for a DVD on TV, then interlaced all the way!! If you are targeting a computer then you have to deinterlace sometime.. :)

~MiSfit
Targetting only a computer monitor.

Thanks for the reply.

henryho_hk

9th October 2006, 02:47

Are laptop harddrives able to sustain such DV captures (in terms of durability, not speed)?

For the recording mode, most PC deinterlacers are converting 30i to 30p also (unless you are using bob deinterlacers which convert to 60p). These "artificial" 30p outputs (yes, there is more than a handful deinterlacers) look definitely different from real 30p recordings but then only your GF can judge which one looks better.

BTW, what is your target x264 bitrate?

Jumbie

9th October 2006, 03:05

Are laptop harddrives able to sustain such DV captures (in terms of durability, not speed)?
Sorry, but I don't understand what you mean by durability.

I have a pretty fast laptop (Core Duo T2500 - 2.0GHz) and a 100 GB (about 60 GB free) 7200 rpm hard drive. Got 3 dropped frames once on an hour-long capture (with windv) that I was testing but I was also doing a lot of other stuff at the same time. She's away now so I haven't tried any longer captures than that yet and I just got this computer recently so haven't done much with it either in terms of recording.

Previously, I used to record to Hi8 tape and then transfer it so 1 hour was the limit on recording time unless I wanted to umm...interrupt to change tapes. I can potentially record now for a few hours if need be.

For the recording mode, most PC deinterlacers are converting 30i to 30p also (unless you are using bob deinterlacers which convert to 60p).

That's what I figured. Therefore, since I was doing that (not using a bob) I'm thinking that recording straight to 30p progressive lets me avoid the deinterlacing step potentially producing better results.

These "artificial" 30p outputs (yes, there is more than a handful deinterlacers) look definitely different from real 30p recordings but then only your GF can judge which one looks better.

Yeah, guess I will just have to test it. So far, I've been happy with my deinterlaced 30p(?) results.

Was just using a simple LeakKernelDeint(order=0) filter in avisynth to achieve that. That throws away one of the fields correct? If that's the case, I was going from 30i to 30p like you said earlier.

According to my manual "A picture recorded in this mode appears clearer" so that would seem to be a major benefit to me. However, does that mean I would have less compressibility and therefore worse results on an encode to the same target file size? If so, that could potentially be a bigger drawback than the gain in clarity and make my end result worse.

BTW, what is your target x264 bitrate?

Anywhere from 1500-2200 kbit/s usually.

Mug Funky

9th October 2006, 08:19

you could (smart)bob in avisynth and encode at a higher rate in x264. don't want to be losing motion now...

9th October 2006, 17:57

Jumbie

11th October 2006, 17:34

you could (smart)bob in avisynth and encode at a higher rate in x264. don't want to be losing motion now...
Can you suggest a good filter that I could try in order to do that?

My footage is bottom field first.

Would something like

assumebff()
separatefields()
a=selectodd()
b=selecteven()
interleave(b,a)
weave()

be a good idea or is there a better method? I'm not very experienced with deinterlacing and don't have a ton of time to do much research with school and all.

Thanks.

2Bdecided

11th October 2006, 19:14

That won't do anything.

Just use...

bob()

That's it! (If it's jumpy, then you need assumebff too!)

Later on, you can replace bob() with SOMETHINGELSEbobORdeint() if bob() isn't good enough.

Cheers,
David.

Jumbie

11th October 2006, 19:16

lol, you're right. It won't. Silly me. I "stole" it from another script that had a different purpose.

Wish avisynth.org wasn't down so that I could look up some of the commands.

Thanks for the advice David.

Mtz

30th March 2007, 21:07

Well - I prefer to record at 25p (PAL) in 16:9 mode (anamorphic). No more hassle with deinterlacing of any kind, and conversion to both, MPEG-4 and MPEG-2, is easy. Interlaced video sucks, if you ask me...

bb
I was thinking at the same thing with my Sony DCRHC96 when I discovered the 25p mode option.
But now I have some dialogue with author of Gspot because the program reported the DV file as interlaced, autogk after analizing reported as progressive and I am a bit confused also after reading the manual which have the same description as Jumbie posted.
I put some test files here: http://www.sendspace.com/file/1sejtj with a small sample when moving the camera left-right and I made some encodes.
From your experience with 25p (even it is a "pseudo-progressive") with CCE can you share some settings about the latest CCE? I said the latest because it can also output bottom field first and is no needing for converting to TTF.

enjoy,
Mtz

stegre

31st March 2007, 08:31

Blue_MiSfit

2nd April 2007, 06:56

One important thing to think about is whether or not your camera has progressive CCDs. If it has interlaced CCDs (like most consumer or low end prosumer cameras), then you will likely get better results recording interlaced, and deinterlacing on your PC. Software based solutions will likely give better results than the hardware deinterlacer in a camera that had interlaced CCDs but can record progressive.

Pretty pretty please use something better than a dumb bob()! TDeint(mode=1) to me does a fantastic job of bobbing, and if you need same rate deinterlacing, it's hard to find something with a better quality/time ratio. Not scripture or anything, just my personal preference, and surely better than bob() :)

~MiSfit

zambelli

5th April 2007, 04:44

Hi Mtz, as I mentioned in the other post (http://forum.doom9.org/showthread.php?p=977417#post977417), I ended up here too, while reading up to try understand this. Maybe bb or one of the other DV experts here can help shed light Sony's system is - is it actually progressive or not? This is the additional description I've found besides what's already been quoted from the manual above:
I'm gonna merge your reply from the other thread here, as it's more appropriate to continue the discussion here:
I was thinking about it too, and I totally agree with above quoted part of your explanation. But as far as the exposure time itself - using PAL numbers here - I think one thing you could not get is a "1/25th sec exposure".

If his camera had manual exposure time control, the longest exposure time possible in interlaced mode would be 1/50th of a sec, and Im guessing the same remains true in his camera's progressive mode. I think the "mechanical shutter" device mentioned is a reference to a mechanical shutter (duh), which, in progressive mode, blacks out any additional light from coming in on every second field. Then, during the second field time, the camera proceeds to clock out the other half of the scan lines from the CCD, which, as you say, still contain the same "persistent" image as the prior field. No temporal difference & hence no interlace artifacts.

However, this method could, and probably often is, used with very short exposure times, which is why one of the advertised features is that it's "you can reduce image blur when recording moving pictures on tapes, intended for import to your computer as still images. This is especially useful for analyzing high-speed action, such as sport scenes".

What this means is that with short exposure times, this "shutter system" approximates true progressive - when it too is set to a fast shutter speed - almost exactly. In both systems each frame would be a series of clear frozen snapshots representing, say, the first 1/1000th of sec of each 1/25 sec interval of real time. This would be good for creating stills for analyzing high speed action.

At the other end of the spectrum is one of the only ways it differs from a film camera or true progressive - something I'm guessing you can't do with his camera's system. With real film or real progressive, if you wanted to, you could set the longest possible exposure time to be 1/25th sec. Then fast moving objects would be integrated over that time into a blur, but the "end of one blur would be the beginning of the next" on consecutive frames, which might be desirable as a film effect for a fast moving object. With this system, at the max shutter speed of 1/50th sec, you'd get a 50th sec of "blur" and then a lapse in time. The following frame's "blur" would start at another position - there'd be a gap in between. And that may tie in with their statement about how "a moving subject may appear awkward".

All the above is speculation, but besides basing it on the manufacturer's statements I mentioned, I did take a close look at DJ's "progressive" samples, and they individually appear to be quite clear - as if short exposure times - then there's a big "distance" jump to the next frame. The frames do not look like an exposure that was averaged over the entire 1/25th sec.

So you're saying:
If his camera had manual exposure time control, the longest exposure time possible in interlaced mode would be 1/50th of a sec, and Im guessing the same remains true in his camera's progressive mode. I think the "mechanical shutter" device mentioned is a reference to a mechanical shutter (duh), which, in progressive mode, blacks out any additional light from coming in on every second field. Then, during the second field time, the camera proceeds to clock out the other half of the scan lines from the CCD, which, as you say, still contain the same "persistent" image as the prior field. No temporal difference & hence no interlace artifacts.
I see your point - if the mechanical shutter speed was 1/25 but the CCD continued to capture images every 1/50th of a second, the 2nd image captured would always be a blur. However, even your theory that the shutter is open only 1/50 secs while the CCD continues to process the same image during the 2nd 1/50th - that requires a change in the CCD behavior, which seems more than just a "mechanical shutter system". The CCD would need to be able to buffer the first image and continue processing it.

One thing I'm pretty sure about is that the progressive image sequence is encoded to DV using a PsF method - the image itself is progressive, but encoded as interlaced, with the fields representing segments of the progressive frame rather than actual fields with a temporal difference.

JohnnyMalaria

5th April 2007, 16:01

One thing I'm pretty sure about is that the progressive image sequence is encoded to DV using a PsF method - the image itself is progressive, but encoded as interlaced, with the fields representing segments of the progressive frame rather than actual fields with a temporal difference.

The DV specification is neutral regarding interlaced vs. progressive. i.e., the video image is always encoded as a 720 x 576 (or 720 x 480) image with the same arrangement, irrespective of interlaced vs. progressive.

However, the DV specification defines three flags that are encoded in the VAUX pack data which tell DV decoding hardware/software how to interpret the video image.

The flags are called FF, FS and IL:

FF: Frame/Field flag

FF indicates whether both fields are output in order or only one of them is output twice during one frame period.
0 = Only one of two fields is output twice
1 = Both fields are output in order

FS: First/Second flag

FS indicates a field which should be output during field 1 period.

0 = Field 2 is output
1 = Field 1 is output

FF.........FS.........Output field
1...........1...........Field 1 and field 2 are output in this order
1...........0...........Field 2 and field 1 are output in this order
0...........1...........Field 1 is output twice
0...........0...........Field 2 is output twice

IL: Interlace flag

IL indicates whether the data of two fields which construct one frame are interlaced or non-interlaced.

0 = Non-interlaced
1 = Interlaced or unrecognized

(From IEC 61834-4:1998(E), "Recording – Helical-scan digital video cassette recording system using 6,35 mm magnetic tape for consumer use (525-60, 625-50, 1125-60 and 1250-50 systems) – Part 4: Pack header table and contents" )

smok3

5th April 2007, 16:35

yes, it is most likely just one dv encoding method, and that is: interlaced, but:

DV uses intraframe compression: Each compressed frame depends entirely on itself, and not on any data from preceding or following frames. However, it also uses adaptive interfield compression; if the compressor detects little difference between the two interlaced fields of a frame, it will compress them together, freeing up some of the "bit budget" to allow for higher overall quality. In theory, this means that static areas of images will be more accurately represented than areas with a lot of motion; in practice, this can sometimes be observed as a slight degree of "blockiness" in the immediate vicinity of moving objects, as discussed below.

from http://www.adamwilt.com/DV-FAQ-tech.html

FredThompson

18th April 2007, 11:35

Why don't you record as interlaced then do a proper conversion to double the frame rate. You'll get 50 or 60 fps depending on the source, eliminate the deinterlacing step and reatin the motion sensitivity of interlaced source?

Alternately, playback using ffdshow and a deinterlacer like kerneldeint.

For best results, to be quite honest, you should really give us some sample videos of bodies in motion so we can help evaluate your options.

stegre

19th April 2007, 08:04

...the DV specification defines three flags that are encoded in the VAUX pack data which tell DV decoding hardware/software how to interpret the video image... IL indicates whether the data of two fields which construct one frame are interlaced or non-interlaced....
"Yes, I do know this, and since the question originally related to what GSpot was reporting, I analyzed the samples in detail. From my own post: (http://forum.doom9.org/showthread.php?p=977417#post977417)...I did take your Type 2 DV AVI files and "de-muxed" them back to a RAW DV stream, which is probably bit for bit identical to what came out of your camera. I then confirmed that "bit 4 of PC3 of the VAUX source control pack" (not too get too technical or anything ;) was indeed set to "1" - which indicates "interlaced" - on both your samples....Now, regarding the merits of "camera hardware" vs. software de-interlacing...... Software based solutions will likely give better results than the hardware deinterlacer in a camera that had interlaced CCDs but can record progressive.In answer to this and some other posts, I had done some experiments and documented them with screenshots, and even made a quickie webpage of all this a couple weeks ago, though I never got around to posting it. The page compares the camera's hardware "mechanical shutter system" to various software based solutions.

Now that this thread is active again, it reminded me - so I just added a few finishing touch or two today - including making the original files available as requested. You'll it find all that right here:

http://www.ftyps.com/unrelated/interlace/

Disclaimer: Don't infer that I necessarily think "still shot quality" is the "end-all be-all" determination of what's best (though it's certainly something very useful). For example, I imagine a full contiguous 25th sec exposures which are "blurry" (but with no interlace artifacts) yet "temporally gapless" from one to the next - something that should be possible with a true progressive system - might be desirable for a certain type of film effect; I certainly acknowledge that possibility. The webpage here only analyzes the quality of "stills" made with the "mechanical shutter system" vs. stills made with a variety de-interlacing software - including sophisticated ones like TDeint - and the shutter system does seems to win out in this regard, anyway.

As I've mentioned, I think what this system does is takes a very short exposure (prob much less than a 50th of a sec), then spends the remainder of that 50th of a sec clocking out the bottom lines. It then probably mechanically blocks light during the next 50th of a sec and clocks out the top lines from the same original image, still "latent" in the CCD. You'd then have no interlace artifacts, because the odd & even lines are from the same image taken at a single point in time. Finally, 1/25th of a sec later, it performs the next fast exposure. So it makes great stills, but if my theory is true one could argue that "half the information (temporally) is lost".

This is all speculation, though - although it's a theory consistent with the stills shown on the web page above as well as with what I've seen from the videos themselves. Links to the videos are there for the taking for anyone who is interested in doing further analysis.

zambelli

19th April 2007, 23:29

, regarding the merits of "camera hardware" vs. software de-interlacing...In answer to this and some other posts, I had done some experiments and documented them with screenshots, and even made a quickie webpage of all this a couple weeks ago, though I never got around to posting it. The page compares the camera's hardware "mechanical shutter system" to various software based solutions.
I think those "mechanical shutter" images are pretty impressive. No sign of aliasing and the images appear to have full resolution. I'm really curious how the Sony camera is accomplishing that. Have you asked around on any other forums?

WorBry

21st April 2007, 09:04

Maybe its something akin to the 'pseudo-progressive' (Pro Cinema) 16:9 mode available on the Panasonic NV-GS400 and 500 cams in which (as I understand) both fields are shot simultaneously (and presumably weaved) to produce a full frame; thus, technically the output remains interlaced (50i) but appears to be progressive (pseudo 25p). Just acquired a PAL GS400 and I have to say I'm quite impressed; not nearly as jittery as I thought it would be. This said, it appears that other image adjustments (nearly said tricks) are also involved in creating the overall 'film effect' (the gamma curve for one) and I'm wondering if there might be a little motion blur in there also. Those who have done comparative tests conclude that there is no perceptible loss in vertical resolution, as theoretically might be expected.

http://media.pana3ccduser.com/wiki/index.php?4%3A3%2016%3A9%20Procinema

stegre

25th April 2007, 16:48

After a lot of Googling - I even went so far as to use Google's new patent search and actually attempted to read the text of some of them - I think I may actually understand how it works. I now believe my "speculation" above, while it has some correct ideas, is not particularly accurate. When I have some time this weekend I'll try to explain the way I now think it works, maybe with a few helpful diagrams.

BTW, Zambelli, remember a response I made to a statement of yours where where I "argued" that for PAL field based video, the maximum "exposure time" couldn't possibly exceed 1/50th sec - it seemed almost obvious me that that was "illogical". Turns out that statement is actually incorrect, and to some extent the common implementation of this mechanical shutter system actually depends on the incorrectness of that statement. Anyway, more later, soon as I get around to it.

zambelli

25th April 2007, 19:25

26th April 2007, 07:54

I asked one of our resident video technical experts on his opinion too - and he had an interesting theory on how this Sony implementation might work. PM me your e-mail address and I'll forward you the thread.
Why don't you post it here?

bb

stegre

3rd May 2007, 08:26

I did spend a lot of time making some nice diagrams (if I do say so myself ;) though I haven't written any description yet, which they certainly require. The preliminary diagrams are here:

http://www.headbands.com/gspot/misc/dvinterlace/

But in a nutshell, the key things to know are:

1. There are several types of CCD chips, all of my diagrams refer to "Interline Transfer" CCD's. These have the capability of picking a point in time very quickly (e.g. ~1ms) performing a near simultaneous transfer the current state of all the pixels to these "store and hold shift registers". This CCD type is common now - used on most cameras, I believe. The quick transfer gives you the ability to essentially set an "exposure time" without any real shutter, mechanical or otherwise.

2. There are two operational modes commonly used, "Field Integration" and "Frame Integration" - some cameras actually offer a choice. I won't go into it all that right here, but they are quite different.

3. If "Frame Integration" available, the exposure time for a field can actually exceed the time of the field - it can be as long as a whole frame. When this happens, the odd and even field exposures overlap in time, though they are "skewed" by one field time - so the result normal "Frame Integration" appears very similar to "Field Integration". Both appear like traditional interlaced content.

4. But once you have that Frame Integration overlap "available", you can add a "mechanical shutter" to the system. That's where it gets interesting. While it's open both fields get exposed - but at the same instant and for the same duration. The "skew" is now gone. And that's the only change that's needed; the rest of the logic runs the same as in "regular" Frame Integration mode.

==========
Note: In the first diagram, the first frame is correct and but the second - labeled "Without Line Averaging" - is there for future discussion. It i s not a normal mode - ignore it. The other two diagrams show identical and presumably accurate pairs frames.

Ref:
1. A description of the two Integration modes, and the inspiration for my diagrams [PDF] (http://www.pulnix.com/Datasheets/Obsolete/TM-75_76.pdf).

2. Obscure reference to the two modes and how a mechanical shutter is added (http://www.damtp.cam.ac.uk/lab/digimage/cameras.htm).