PAL to NTSC, Vegas is lousy! A LONG educational post with some questions... (farss and mbryant check this one!)

GTakacs wrote on 6/21/2004, 2:50 PM
Hi all, this is my very first post.

I have been lurking around for a while and searched a lot of archive posts relating to PAL and NTSC conversion.

I live in the US and I have friends and family living in Europe. I volunteered to convert all my father's old VHS8 tapes to DVD. These are all old analog PAL footage. Since I want it to play both in the US and in Europe I have decided to convert it to NTSC and do my final authoring in NTSC for all my project.

The reason is because PAL players in Europe will play NTSC DVDs just fine (as long as there is no region code) while NTSC players will not play PAL DVDs for the most part (mine just says, improper video format for this region) and choke on it. I have also read that PAL to NTSC conversion is better as there is more initial resolution to convert from, giving better ultimate result.

My problem, as so many other people's is the PAL to NTSC conversion. With interlaced footage it is not as easy as it seems. And de-interlacing loses half the resolution right off the bat, and while it might not matter that much with some old VHS8 footage, I would like to retain as much of the already poor quality as possible without introducing more artifacts.

I have searched previous posts and opinions on this matter and it seems like that some people swear by the simple solution of using Vegas to do the rendering and call it a day.

Some have used other pieces of free softwares and plugins to get better results, but it seems just way too complicated and the people in the first camp (just use Vegas and live with it) seem to claim that their results are just as good as the others.

Well I decided to do my own test.

I have captured some analog PAL footage from my old VHS8 camcorder using my ATI All-in-wonder Radeon 9600XT video card. One drawback of using this card is that it can only capture in MPEG2. Well it can capture in AVI standard as well, but there is no decent codec that I have found that will do a reasonable job with keeping file size manageable while keeping quality decent. The card does an OK job capturing into MPEG2 with 8Mbit VBR, it is nothing stellar, but it is definetely fine for some old analog footage. The only drawback is the need of another conversion, however that is not that big of a problem since I have to convert this footage from PAL to NTSC anyway.

So here I am with some captured PAL MPEG2 footage that I would like to have in NTSC DV codec so I can edit it and render into a final cut.

I have tried to use Vegas to just do the conversion, so I set up a new project with NTSC DV as the target and loaded in my PAL footage. Then I made screen captures of the preview window with all 3 de-interlace methods (none,blend,interpolate). Also, the original PAL footage was 720x576x25i with top frame first. I wanted NTSC DV with 720x480x29.7i with bottom frame first.

Here are the samples of my captures. The item in the picture is supposed to be a waterskier going towards the right at fairly high speed. This helped me to confirm the interlacing and which frame is top and which is bottom. I could confirm that the PAL footage I have captured from the analog source was indeed top frame first.

Original interlaced PAL footage

Vegas NTSC Iinterlaced with no de-interlacing

Vegas NTSC interlaced with blend de-interlacing

Vegas NTSC interlaced with interpolate de-interlacing:

The interesting thing is that all 3 de-interlacing methods produced identical results. They are all the same, so I am not sure how is that possible. It might be that de-interlacing is not applied to the preview screen, or I have scewed something up, but nevertheless they all look the same. If someone could confirm this, that would be great!

However, what is even worse than that, is that the interlacing is completely gone! You can see how the resampling totally ruined the footage, I bet that this video would experience some serious artifacting on TV. This is an extreme of interlacing (fast moving object) but it just looks like the interlacing or lack of de-interlacing completely ruined the footage. This is not acceptable.

If anyone has some other response, opinion, solution, feel free to add to it, I am all ears! I will try tonight and actually render the footage to NTSC with all 3 pre-sets and look back the final rendered NTSC DV footage to see if it's just a preview anomaly or not.

I have abandoned the option of using some open source freeware application with several other plug-ins or DLLs to install as a solution to my issue. It just didn't seem simple enough for my taste. I have decided to try using Canopus ProCoder and see what kind of results it would produce.

Here is the render after putting it through Canopus ProCoder


Since this is the only process that produced properly interlaced results that are actually interlaced and not ruined, I counted the lines from border to border, and they are in fact field swapped from the original, like they are supposed to be in NTSC DV footage, bottom field first. Needless to say, this is the only footage that is worth looking at.

Whoever said Vegas does a decent job of deinterlacing and resampling PAL to NTSC was wrong. It looks like poop compared to the Canopus result. I might be wrong and didn't have the rendering set up right, but unfortunately I think I am right on this one.

And the floor is open for suggestions!

Also, what will I want to render it as if I want to produce a DVD off of it? Does NTSC DVD use bottom or top frame first in interlace?

Comments

johnmeyer wrote on 6/21/2004, 2:58 PM
I don't know much about PAL to NTSC (I've only done it a few times). However, I would think you should be able to capture in DV format rather than MPEG2, either through a camcorder, or with some sort of update to your ATI card. Capturing in MPEG2 gets you off to a bad start (quality wise), and forces you to do another step to get it into AVI format so you can edit it.

Also, I don't understand why you would want to deinterlace. Normally you only deinterlace if you are certain that the output will only appear on non-interlaced monitors. Someone else can jump in and tell me if I am missing something here.
BrianStanding wrote on 6/21/2004, 3:00 PM
I think you would do far, far better to capture into PAL DV, rather than MPEG-2. MPEG-2 is a lossy, format that is not designed to be recompressed.

A cheap PAL DV or D8 camcorder with analog inputs (or does the VAT prevent such beasts?) and a firewire connection will probably do a better job than the ATI card. Canopus also sells some reasonably priced analog to DV converters that may be PAL-compatible.

I've had decent luck converting from PAL DiVX to NTSC DV with some stuff with a high bitrate in Vegas. There's some artifacting on high-motion/FX sequences, but otherwise looks pretty clear.
farss wrote on 6/21/2004, 3:06 PM
So much to cover here.
Firstly both NTSC and PAL are lower field first, period.
Secondly, on a computer monitor you can only preview in progressive so yes you may endup seeing horrible interlace artifacts. Converting to progressive to try and get around that only introduces more problems so don't even think about it. If the output from Procoder shows no interlace artifacts then either you're driving it wrong or it's a crock, I suspect the former.
Thirdly capturing using mpeg encoding from old VHS no matter what you do with it later is a formula for very poor results. VHS and old stuff inparticular is going to have a lot of noise and jitter and any mpeg encoder is going to make that much worse.
So your first step should be to get a decent capture, I use an ADVC-300 so I can a) get time base correction to remove as much jitter as possible and b) use Dynamic Noise Reduction (a temporal process) to get rid of some of the noise.
Also changing the de-interlacing method in Vegas will have no effect if you're not de-interlacing.
So first up before you even worry about the PAL to NTSC conversion issues you need a better way to capture. Once you get to that stage you will always get some quality loss going to NTSC, 'tis the nature of the beast. I seriously doubt you'll see anything if you started out with VHS though. I've done many NTSC DVDs using the simplest method of encoding from the TL straight to NTSC mpeg-2. At that point you MAY find one de-interlacing method gives better results than the other, I'd strngly suggest running some tests of your own on short bits of footage.

P.S. If you don't have an ADVC-300, a Digital 8 camera is nearly as good and much better than a DV camera for doing the conversion.
GTakacs wrote on 6/21/2004, 4:20 PM
I understand that getting the footage into my computer as MPEG2 is not the best way of importing it. But since I have to convert the footage from PAL to NTSC anyway, I think I can do that and the conversion to DV at the same time.

As for the deinterlace, if you would look at the pictures that I have posted from Vegas you would understand why you would want to de-interlace. Since Vegas resamples the entire video frame (both A and B fields interlaced together) it ends up with a result that is resampled from 576 to 480 and it screwes up the interlaced fields as you can see. Just take a look at the pictures I have posted earlier.

But since Canopus does a phenomenal job in real time (I have a P4 3.0GHz with 1GB RAM) in first breaking up each frame to A and B field then resampling them and weaving the results back together for a true interlaced frame, there is no need to de-interlace wit this method.

I hope I was more clear this time....
GTakacs wrote on 6/21/2004, 4:23 PM
First, I am not going to spend more money on a DV camcorder that I would never use other than this one time to get all my old footage, I live in the US, buying a PAL camcorder makes no sense what-so-ever.

I agree that MPEG2 is not the ideal digitizing form, but see my earlier response above, it does not really matter when it comes to the NTSC to PAL convert, that is where the real issue is.
erratic wrote on 6/21/2004, 4:26 PM
ATI Radeon cards capture upper field first, so you're not only converting your PAL MPEG-2 file to DV NTSC, you're also changing the field order at the same time. I've had good results converting interlaced PAL AVI (tff) to interlaced NTSC (also tff) with Vegas, but changing the field order got rather blurry.
farss wrote on 6/21/2004, 4:39 PM
Good point, camcorders only come in PAL or NTSC, good argument for ADVC-300 or 100, they handle both.
Beside the issue of it not being the best way to capture you need to get your heard around the idea that what you see on the PC monitor should have interlace artifacts on anything with fats motion. If you're working with interlaced video and you don't see them I'd suggest something is wrong!
Using an external monitor is the ONLY way to truly see how your output will look. If as has been suggested your video card is capturing with the wrong field order then that'll cause other issues however I owuld have though Vegas can reverse the field order without any quality loss however things may get messy coming from mpeg. Mpeg I think always does a field merge, it's only flags that tell the decoder to split the fields back out into frames.
erratic wrote on 6/21/2004, 4:45 PM
Well, farss, I can assure you that the Radeon cards capture upper field first, but this is not a 'wrong' field order. Vegas has no problem converting tff pal avi to tff ntsc. But as I wrote in another thread having tff source video and converting to bff dv isn't so good.
John_Cline wrote on 6/21/2004, 4:46 PM
I regularly use the Canopus ProCoder software to do NTSC <> PAL conversions. It is the best software solution I have found and the results are impeccable.

John
GTakacs wrote on 6/21/2004, 6:04 PM
Firstly, PAL footage that is non-DV is actually upper field first. I have verified it on that waterski footage, the skier was more to the left on the odd rows than on the even ones. And given that the guy was going right, it is clear that the odd rows were recorded first. On PAL DV you are correct that lower field first is the standard. Just check your presets in Vegas, you will see that PAL DV is lower field first, but PAL Standard is upper field first, just as I have confirmed with my own footage.

Secondly, I fully understand that on my monitor, in preview mode I will see one progressive frame of the interlaced A and B fields and I will see the jaggies. I am SUPPOSED to see them. But I am supposed to see two very distinct patterns.

I have made a new example shot, that should illustrate the issue better.

http://gregtakacs.com/PAL_NTSC/interlace.jpg

The top line is the original footage right off the tape, as it came in. The second line is converted to NTSC using Vegas by setting the project to be NTSC DV and just rendering the PAL input file straight out. The third line is what Canopus rendered after I told it that the source was PAL video (top first interlaced) and the destination is NTSC DV (botton first interlaced)

The first column is the capture of the top right corner of the rendered frame. The second column is only field A (top) by itself, the third is field B (bottom) by itself, the third is field A twice with an offset of one vertical pixel and the last column is field B twice with one vetical offset.

You can see how manually de-interlacing the image on the PAL source the skier is further to the right on the B frame, hence the A frame was captured first, which is the top frame. This is proof that the PAL footage I have is indeed top field first.

You can also see how Vegas blatantly ignores interlacing and the A and B frame are a mixture of the original PAL A and B frame due to downsampling. Sometimes the original A row sometimes a B row happened to fall onto the NTSC A and B rows respectively. Playing this footake will reveal serious artifacts. The way to fix it is by elliminating one of the field entirely making it a 240 horizontal resolution progressive footage essentially.

Another solution is to use Canopus. You can see how canopus retained perfectly distinct separate interlaced fields with good resolution. You can also see that on the Canopus render the skier is further to the right on the A frame which means that the B frame was recorded first, ie. it is a bottom field first, just as the NTSC DV standard requires it.

Please take a look at my image, I hope it will make it crystal clear what I am talking about when I say Vegas ruins the interlacing in its entirety. The only solution is to deinterlace the image and lose half horizontal resolution.

And as I have pointed out in my previous answers, I understand that capturing in MPEG2 is not the best option, but I can guarantee you that you lose a lot more in the PAL to NTSC conversion using Vegas than I do by using MPEG2 as my initial capture.
GTakacs wrote on 6/21/2004, 6:07 PM
I think people still don't get the fact that Vegas does a terrible job converting, I hope my little post above does help to point that out.
GTakacs wrote on 6/21/2004, 6:09 PM
My point is that once you have PAL footage in your PC, regardless how you got it there in what format, converting it to NTSC using Vegas will ruin interlaced video in its entirety.
BrianStanding wrote on 6/21/2004, 7:57 PM
MY point is that I have done successful PAL to NTSC conversions using Vegas. I just don't use MPEG-2 source material.

Beauty, of course , is in the eye of the beholder. What I consider fine, you may well consider unacceptable.
Spot|DSE wrote on 6/21/2004, 8:49 PM
Using MPEG 2 as a comparison source for a product that is openly clear about not being optimized for is somewhat silly. Might as well complain that it's not a good apples to oranges.
Vegas is a DV app first and foremost. It does not cut mpeg nor wmv, nor other low resolution formats well at all. While it does indeed cut mpeg, this could just as easily be shut off because in this sort of event, it makes Vegas look bad when used improperly.
We regularly convert PAL DV to NTSC, and will show going both directions when I arrive in Australia next month for the VASST tour there. The interlacing issue is moot at that point. I'm sure there are freeware and shareware tools that do this much better, because there are tools optimized for editing MPEG and converting it to other formats. But Vegas isn't one of them.
NTSC or PAL, expecting good results on the outcome is wishful thinking when the source is MPEG, unless its a short MPEG source, which I doubt we're talking about here. Long MPEG doesn't edit well on much of anything period.
GTakacs wrote on 6/21/2004, 10:25 PM
Just look at this picture:

http://gregtakacs.com/PAL_NTSC/interlace.jpg

And everything will become crystal clear as what I am talking about. It has NOTHING to do with the fact that my original is MPEG2. I can convert my PAL MPEG2 to PAL DV and it would still not change the fact that the interlace would look PERFECT as long as I keep it in PAL (regardless of format) and as soon as I try to convert to NTSC the interlacing goes right out the window using Vegas. Why is this so hard to comprehend?

It seems like people completely ignore what my initial post was about and keep hammering on the fact that I use MPEG2 as my primary source. For a test I converted my MPEG2 source to PAL DV using Vegas, and the resulting PAL footage looks fine, everything is perfectly interlaced just like the original MPEG2. I have to add that the rendered PAL DV is bottom first as it should be, so if one thing Vegas does properly is to perform the field swapping (my original footage was top first). Edit: It seems like Vegas does not do this one right either:See This

But as soon as I try to render my project as NTSC DV, I get the same resampling artifacts as I get with MPEG2. Interestingly enough Canopus ProCoder had no problems dealing with my original MPEG and performing the downsampling, interpolating new frames and swapping my frames to be bottom first from top first.

Granted that the Canopus software costs as much as Vegas in its entirery, maybe this is the reason..... Maybe I was expecting too much from Vegas to do on its own, it clearly fails in the PAL to NTSC conversion department.

Again, PLEASE look at my picture and discuss that, not the fact that I use mpeg2 compressed footage as my original.

http://gregtakacs.com/PAL_NTSC/interlace.jpg

With this many pros around, I can't believe that noone accepts the fact that Vegas PAL to NTSC conversion does not work. Only "erratic" posted something that was slightly agreeing with me.
mbryant wrote on 6/22/2004, 1:02 AM
GTakacs,

As I was mentioned in the initial post I thought I should reply, but I'm afraid I am just an amateur and can’t add much over what has already been said.

I routinely convert PAL DV footage to NTSC-MPEG using Vegas, and while I do notice some loss of quality, in my eyes this is minor and the results are very acceptable. I’ve also used ProCoder Express, and this worked fine too – though I found the Vegas conversions slightly better.

I understand that you are capturing in MPEG rather than DV, and you’ve seen all the mails suggesting DV would be better, Vegas isn’t so good at editing MPEG, etc. I understand your point and frustration that you can convert the PAL-MPEG to PAL DV and it is OK, it is only when you convert to NTSC you have the problem, so the problem seems to be in the conversion and not the MPEG capture. But that’s where I lose you; I’ve looked at your images but I don’t understand interlacing enough to understand what is happening.

Mark
farss wrote on 6/22/2004, 4:18 AM
You might like to try turing on Reduce Interlace Flicker in media properties. I've done many PAL to NTSC conversions and they looked way better than that. Seems to me something else is going wrong, have you done any resizing of the frame?
I'm not certain of this however, also if the field order was wrong to start with the the de-interlacing could be getting seriously screwed up.
We've had the results of my PAL to NTSC conversions up on a studio monitor and they looks as good as any conversion that started out as DV.
PeterWright wrote on 6/22/2004, 4:45 AM
When you convert PAL MPEG2 to DV, it may appear to look alright, but you have overlooked what many have already told you.

If you use MPEG2 as your original, you are choosing a highly compressed format aimed more at display than editing. Even if you convert this to DV, it is starting with a lack of digital information,.

Start again. Capture analogue through firewire rather than your graphics card.

You need either an A>D converter (e.g/. Canopus) or a DV/Dig8 camcorder.

Do that, then convert to NTSC, THEN tell us Vegas is a failure. Give it a chance in its native format.

Spot|DSE wrote on 6/22/2004, 4:46 AM
You might also insert Median as a noise reducer.
If you're gonna post that Vegas is lousy, and then do a non-standard convert, with non-standard media, with probably a junk converter, with god only knows what inserted by the converter....you can't blame Vegas, or at least claim that Vegas is lousy. Seems to me you missed your own point. Many of us convert in both directions, you've read that now. There is obviously some problem with your stream. That doesn't make it Vegas' fault. Regardless of quality.
Also, ALWAYS expect a quality hit whether it's seen or not when transcoding.
Veggie_Dave wrote on 6/22/2004, 6:06 AM
Peter Wright: Do that, then convert to NTSC, THEN tell us Vegas is a failure. Give it a chance in its native format.

Okay, that's exactly what I did - Vegas' results were appalling and not even remotely acceptable! If your footage is slow moving or even static, then it might be acceptable ... but for those of us whose footage most definitely isn't slow or static then the results can best be described as a joke, especially for an NLE that's supposedly a professional product

I love Vegas, however I have no choice but to either use Procoder or send the finished footage for mechanical conversion because the results in Vegas are simply &(*&^%^!
GTakacs wrote on 6/22/2004, 6:09 AM
Did any of you looked at the picture that I painstakingly assembled in Photoshop indicating the Vegas interlace conversion?

It is not at all difficult to comprehend, yet my whole complaint has fallen onto deaf ears.

I can't strees enough that it has NOTHING to do with me starting with an interlaced PAL MPEG2 footage. I could take a single still frame AVI file that shows interlacing and use that as my initial test (well except for the inter frame interpolation that actually needs more than one frames, but let's not go there).

I can see that NONE of you guys actually took the time to capture Vegas output and analyze under the microscope (by cropping and enlarging interlaced footage that shows high speed motion suseptible to resampling artifacts). All of the professional's attitude seems to be "if it looks good on the studio monitor, that's good enough for me". Since I am an engineer by profession, not a video guru, I have gone down a different road to verify the working of the program.

I am going to try to explain one last time my issue/problem with Vegas, and PLEASE concentrate on this and pretend that I have started with a PAL DV footage, believe me, IT DOES NOT MATTER!

http://gregtakacs.com/PAL_NTSC/interlace.jpg

I have split up the interlaced video into the two fields. The two fields in theory are two progressive half resolution images, and they should show two distinct yet sharp images of the subject matter. You can see how the original PAL footage does just that. There is the waterskier traveling from left to right, and the only differerence between field A and field B is that Field B was captured 1/50th of a second later and the skier has already moved a bit to the right, which is expected. So while you see the haircomb effect of the interlacing on the interlaced video when viewed as a progressive frame (A+B) they completely vanish as soon as you take it apart to its field elements.

Looking at the Canopus ProCoder rendered NTSC footage, you can see the exact same thing happening. It is resized, you can see that the skier is a couple of pixels shorter, it is not as clean as the original frame, because it is an interpolated inter-frame, you can clearly see how the skier has double heads, so Canopus only interpolates the two frames and it does not use motion vectors, that would be the ultimate best solution, but let's not be that picky. You can see that looking at the Canopus footage, splitting up the two fields, you still get two distinct instances of the skier, however on this one it is further travelled to the right on the top field, so it is a bottom field footage. But you still see that both field A and field B look perfectly still, no interlacing artifacts on them.

Now if you look at the Vegas footage, you can see how even after de-interlacing the footage by splitting out the A and B frames and doubling them on top of one another (A+A and B+B respectively), you can see that field A and field B both have a fair share of the original field A and field B in them, hence the jaggies.

If I would capture a single frame of the PAL footage, and resize the image in Photoshop from 576 high to 480 high using some kind of interpolation method, and I would take out all the odd lines and make them into a picture and use all the even lines to make another one, I would experience the exact same thing that Vegas is doing. Vegas just resamples the entire frame as a progressive frame, completely oblivious to the interlacing.

I will try different de-interlacing methods in Vegas itself to see if I can get a better result, like I said before, it will definetely be better than what it is producing in interlace mode, but it will lose half the vertical resolution, just as many people have commented on that in prior posts. It will never be as good as Canopus ProCoder, that is for sure.

I have found a solution my problem by using a proper workflow that includes Canopus to convert the PAL to NTSC footage instead of relying on Vegas. I don't think I can be convinced otherwise. I just tried to show it to other people as well that Vegas does miserably with interlaced footage downsampled from PAL to NTSC.

Thank you all for your time! And if you feel you need to discuss this in more detail (NOT the fact that my original is PAL MPEG2 and not DV) feel free to drop me a line at AIM:GTakacs1976 or icq:2636936.
Liam_Vegas wrote on 6/22/2004, 6:15 AM
Just on the PAL to NTSC thing with a camcorder. While travelling recently I found a surprising capability in my Sony PD170 that may also be true for other camcorders. I found that my camcorder can playback a PAL tape (the camera is NTSC). This came in handy when I had to do a dub from a PAL tape someone gave me. I managed to create a DUB of the tape which I can now capture using my existing camcorder as the deck.
donp wrote on 6/22/2004, 6:52 AM
I have only done one PAL to NTSC conversion but I used DV2AVI and then edited a PAL AVI in Vegas (no quality problems) and then converted the PAL DV AVI to an NTSC DV AVI (maye that was an extra step). Then Procoder from the NTSC DV AVI to elemental NTSC m2v and sound. Put that on DVD and it looks just as good as the original PAL DVD I played on the PC player, no fuzzies, artifacts nothing. Mpeg2 is not handled by Vegas well at all as a time line event so I recomend a PAL AVI to start.
farss wrote on 6/22/2004, 7:08 AM
Pretty annoying when people make the effort to reply and you clearly haven't bothered to read what they had to say. Normally I wouldn't care but seeing as how you addressed the topic to me I'm a tad miffed. Yes I did spend some time studying your screen shots, and yes I did make some significant comments, my tests have been compared to systems using motion vectors that cost orders of magnitude more than Vegas or Procoder.

So I've done some more tests using the limited facilities I have at hand. I've created a very fast moving black rectangle on a white background using generated media and rendered that out to PAL DV, interlaced, BFF so there can be no argument over mpeg-2 which I agree MAY be irrelevant.
I'm talking extreme motion here, the rectange takes 10 frames to rotate 360 deg and travel from TL to BR. Switching project properties to NTSC DV, interlaced BFF the results are VERY interesting and do explain what you are seeing.
In Preview in PAL I see the two fields, one with box in one position, the other showing the box in the next position, no surprises there. Switching to NSTC I have THREE fields containing images of the box and in at least two of them I have significant banding, blowing it up in PS looks much like the problem you were showing in your samples.
However I can get rid of it in three mouse clicks, right click track, click Media, and tick "Reduce Interlace Flicker". Banding is GONE, three very nice looking fields. Some degree of motion blur is going to be inevitable without some very fancy kit and most cannot see the difference, remember my test footage had NO motion blur to start with, real life footage would so the effect would be largely masked.

So all I can say is waste your money on Procoder, I'm more than happy using Vegas and ever test I've run gives results comparable to the output from a S&W box. If I was staring with 4:2:2 material then I'd probably run it through a S&W, for DV, Vegas is more than adequate.
I seem to recall suggesting you try this right at the beginning, could have saved a lot of angst.