Laserdisc capture workflow. — Page 2

The first thing to do to improve the capture quality is obviously change the capture card... it must not be a super expensive one; I suggest to avoid USB based (I have one and is really awful); an old and/or used PCI card, that would top quality some years ago, could be a good choice - it will be really cheap nowadays!

Mine is based on the Philips SAA 713x cheapset (could be SAA 7130, 7133, 7134... the bigger the last digit, the better) and it's pretty good.

Anyone consider the laserdisc internal comb filter really basic, compared to today's standard; it can be true, but as many capture card cheapset are old too, sometimes using a s-video out from a laserdisc is not a bad idea, in particular if you use a high quality player - that could have a better comb filter than your capture card.

A simple test is this: capture some footage and let the laserdisc display the time on screen; then look at the digits on the screen; if there are visible rainbow artifacts, the comb filter is not that good...

About the lasedisc players: the first one (CLD-HF9G) is really good, could be included in the top twenty (maybe to ten) laserdisc players of every time; the others are medium quality. If you decide to use the TOOT (two out of three) avisynth plugin, use the HF9G capture as first clip, as it will be the best.

I have to admit I was totally wrong about this in the beginning.  I thought I could just get a raw capture and denoise it to clean up the picture but I've seen that even the best denoiser will not help too much with dot crawl.  It wasn't until I started using a dot crawl/derainbow filter at the top of my filter chain that I started to really see a clean looking picture emerge. I think it helps that I'm using the s-video from my laserdisc player now too.  The comb filter in my usb capture is a 1 line comb filter which is hardly worth using anyway. 

I would like to try the toot filter. I used to have it in my plugin folder several years and many crashes ago.  Does anybody have it that would be willing to share it with me?

Here's the filter set up I'm using now.  I might add the spatial dot crawl filter later too. This script allows me to capture in YUY2 and keep it in the same colorspace all the way down the filter chain without any resizing either.

raw=avisource()

decombed=raw.tcomb(mode=2,fthreshl=5,othreshl=5,fthreshc=220,othreshc=220)

ivtc=decombed.tfm(mode=5,PP=2,clip2=decombed.tdeint()).tdecimate()

vectors=ivtc.mvanalysemulti(refframes=2,pel=2,blksize=16,blksizev=16,overlap=8,idx=1)

denoised=ivtc.mvdegrainmulti(vectors,thsad=200,sadmode=1,idx=2).tweak(sat=1.1).levels(4,1.1,255,0,245)

return(denoised)

Btw.  I tried the little test for rainbowing like you suggested and the only player that doesn't exhibit really bad rainbowing is the HF9G whilst using the s-video connector. It has some but not nearly as much as the other 2 laserdisc players.  They all exhibit rainbowing pretty bad whilst using the composite outputs. Happily, tcomb seems to alleviate most of it.

althor1138 said:

Btw.  I tried the little test for rainbowing like you suggested and the only player that doesn't exhibit really bad rainbowing is the HF9G whilst using the s-video connector. It has some but not nearly as much as the other 2 laserdisc players.  They all exhibit rainbowing pretty bad whilst using the composite outputs. Happily, tcomb seems to alleviate most of it.

If you can try another capture card, probably also your other two laserdisc players might exibit near no rainbowing - as my CLD-D925, that is similar to CLD-606, just a bit above your 505.

P.M. your email address, so I could send you the twooutofthree.dll you need.

I've decided to upgrade my capture card before continuing with this. If anybody can give me some links to some good capture cards I'd appreciate it.  I'm also looking for advice on external comb filters.  This is alien territory for me.  It seems like Faroudja makes some good ones? Apparently there isn't anything that has ever exceeded the X0's comb filter as evidenced here?

EDIT: check page 2/post#43 & 44 if you are interested in that link.

The HLD-X9 S-Video out should be better than HLD-X0, because is 3D and came after two years...

It occurred to me today that I potentially have a good quality external comb filter sitting right in front of me.  I have a 40" Sony Bravia LCD TV that somehow manages to make anything I run through it look good.  I'm assuming this means it has a good comb filter and a deinterlacer at a minimum.  It can output a signal from the scart connector in composite,s-video, or rgb. This means I should be able to use the tv as a comb filter and pass it on to the capture card via an s-video cable.

Can somebody point me to a list of laserdiscs that have test patterns?  I have about 80-100 laserdiscs but I haven't the slightest clue which might have a test pattern that I can use to see what kind of quality the tv comb filter has.

EDIT: nevermind.  It will output from the tv but it doesn't pass it through the comb filter :(.

EDIT2:  I'm just going to bite the bullet and buy a better capture card. It took me a long time to find a saa713x card here in sweden but I think I finally found one.  Surprisingly it's a pinnacle but it has an saa7134 chipset so I put a bid on it.  We'll see in a day or two if I get it.

I've won the bid on the Pinnacle PCTV Stereo.  I also ordered a new card from Amazon in the states.  It's a Diamond Theater 750 HD.  It's supposedly got a better comb filter than the saa713x chips.  It's got a 5 line 2d comb+3d motion adaptive comb.  The saa7134 on the Pinnacle is a 4 line 2d I think.  I've seen some comparison screenshots with the snell & wilcox pattern online and it looks better there but we will have to see how it pans out in real life I suppose.

I've also been looking through the service manual for the cld-99(cld-hf9g) to see if there is any possible way to bypass the 3d y/c of the player.  It looks to me like the RF signal gets fed into a Digital Video Processor IC where it gets converted to digital and clocked. It leaves this chip exactly like this and goes directly into the 3d y/c chip and then into a few other things like the on screen display overlay chip before going to a DAC and leaving the player.  Unless my HF9G is different than the 99, tapping this digital signal between the processor and the 3d y/c chip is the only way to bypass all of the filter chain and sending it directly to a modern comb filter. Unfortunately, this seems to be a bit over my head and involves some other hardware that I don't understand.

Actually looking a bit closer I see that the signal Y's before leaving the Digital Video Processor chip.  One side exits and goes digitally to the 3d Y/C chip and the other hits an internal DAC on the chip and exits on a pin and then goes nowhere? If this is true I could take the analog signal directly after the TBC and not have to worry about a lot of extra equipment. Is there anybody who can confirm this on the CLD-HF9G or CLD-99? I can send the service manual to anybody that wants it. Here's a picture of the chip. I've tried to trace the video signal in a simple way so it's easy to see.

Just an update on my capture situation:  I now have 3 capture devices and I've even built a new pc that should be able to handle capturing  better than my old one.  My plan is to post a few comparison shots between the different capture devices by the end of next week.  I have the usb pinnacle hdtv 340e stick, the pci pinnacle pctv stereo card, and the pci-e ATI theater 750 HD card.

I would also like to tap the signal and try to capture before the laserdisc player has a chance to digitize anything. I'll probably try this on the cld-d505 because it's a cheaper player than the cld-hf9g and my soldering skills are questionable so I might end up destroying the PCB if I am unlucky.  The laser pickup is basically the same as the more expensive machines anyway.  It's just the filtering afterwards that makes the picture look better I think.

I've done a few preliminary tests and the theater 750 hd card looks impressive.  Of course the only thing i can compare it to right now is the uber-cheap pinnacle stick I was using.  It looks like having a decent capture card really does make a huge difference in the quality of the capture lol.

If anybody has any tips or tricks for capturing feel free to share them :). 

The ATI Theater HD 750 has some terrible AGC issues.

Search on the videohelp.com forums for more information on the AGC problems inherent to its chipset.

Ok here's some screenshots:

First off, I think the PCTV card is unusable to me unless somebody can tell me how to get rid of the constant dropped lines? I think the theater 750hd chipset is several notches above the pinnacle chipsets when it comes to y/c separation and rainbowing. If you look at the finer details of the japanese subtitles it's very apparent imo. The usb stick is almost as good as the pinnacle pci card when it comes to rainbowing.  This surprised me as the stick has a 1 line notch comb filter and the pinnacle pci has a saa7134 chipset with a 4 line adaptive comb filter. I haven't noticed any of the problems with agc on the theater 750 like people have been talking about in the forums online.  I'll admit I'm a total noob and maybe my eyes aren't trained to catch it.  As of now though it seems clear to me that the Theater 750HD is the way to go but I am posting these because I want feedback and suggestions so please feel free to tell me what you think.

EDIT: Viewing the images at full resolution reveals a lot more than these resized images btw.

The difference in cross-colour artefacts (rainbows) in the subs in very clear, the Theater card is much better in this respect.

The fact that you get dropouts with the PCI card is surprising, as I always thought these were down to the disc or player. Never heard of the capture device being responsible, but there you go. Possibly a faulty card?

Do you have any discs that contain test patterns? A capture of such a pattern would be a more useful source to compare performance.

It very well could be a faulty card.  I ordered it used on Ebay.  I suppose it could also be a driver issue perhaps?

I have around 100 laserdisc titles but I have no idea if one of them has a test pattern on it.  I'm thinking I probably should break down and order video essentials or something similar.

Well I can't find any test patterns on any discs I own so I have ordered a copy of Video Essentials off of LDDB from the Netherlands.  He said it would be in the mail within the next couple days so maybe by next week I'll be able to post some test patterns with the different cards and players.

Just an update.  I'm still waiting on the video essentials laserdisc to arrive.  The guy from LDDB swears up and down that he mailed it already and that it should be here before the end of the week.

I have been playing around with capturing on the Theater750 and I'm now almost 90% sure that the cld-hf9g does have almost pure composite out. It still passes through the chip that overlays the on screen display but I believe there is no filtering or y/c separation going on. 

If I turn up the y & c noise reduction all of the way and then turn up the Y/C separation all the way to Max and then capture via an s-video cable I get a noise-free picture that still shows grain and a little blur from the 3d y/c separation. There is significant rainbowing however due to the fact that the comb filter is not the best but I haven't noticed much dot crawl problem.

However, If I keep the same settings and capture with the composite out I get a softer picture with more noise but without the rainbowing and no dot crawl.  I'm assuming this is because it is being filtered out with the modern comb filter of the Theater750.

What I've done just playing around is turn off the HF9G comb filter but leave the noise reduction on. Then I did 3 composite caps, 3 s-video caps, toot them respectively and use the luma from the less noisy s-video cap and the chroma without the rainbowing from the composite cap.  It looks pretty nice to my eyes and I don't feel like it needs any more processing other than a proper IVTC.

The difference is hard to spot spatially.  The toot also seems to remove a lot of the rainbowing problem of the s-video caps.  The screenshot below isn't much of an indicator but you can sort of see that the noise is lessened.  I think the best part is that this looks natural compared to using a software denoiser.  There is still grain, nothing looks plasticized, and no dvnr blurring. 

EDIT:  BTW, I should mention that when capturing I didn't have the noise reduction turned up all the way.  I only used the first 2 out of 10 increments. If you turn it up past 7 it starts to blur.

So, it took me 2 or 3 nights but I've finally figured out how to manually IVTC.  I can say with certainty that the JSC/SW must be manually IVTC'ed.  TFM and Telecide fail miserably during the crawl and pan-down.  Once I figured out the order and how to trim the frames properly it's like a clear and stable picture suddenly became visible.  I think my capture card pulls in fields in the wrong order perhaps?  The order on my capture starting with the first frame of the Title card and using separatefields goes like this: 0,1,3,2,5,4,6,7,8,9

althor1138 said:

I think my capture card pulls in fields in the wrong order perhaps?  The order on my capture starting with the first frame of the Title card and using separatefields goes like this: 0,1,3,2,5,4,6,7,8,9

What are these numbers? 

Are you using DoubleWeave().Pulldown() to manually IVTC?

Is there a way to view those images in a larger size?

Right-click, Open in New Tab (Chrome) or View Image (Firefox)

I'm using this method.  It was the easiest for me to understand.  Here's a little snippet of the script I'm using.

cpass1=AVISource().trim()
cpass2=AVISource().trim()
cpass3=AVISource().trim()
tootc=twooutofthree(cpass1,cpass2,cpass3).trim(345,0)
spass1=AVISource().trim()
spass2=AVISource().trim()
spass3=AVISource().trim()
toots=twooutofthree(spass1,spass2,spass3).trim(326,0)
mix=overlay(tootc,toots,mode="luma",opacity=1)
p1=mix.separatefields().selectevery(10,0,1,2,3,4,5,6,9).weave().trim(0,1090)
p2=mix.separatefields().trim(2726,0).selectevery(10,0,1,2,5,6,7,8,9).weave().trim(0,2832)
p3=mix.separatefields().trim(9810,0).selectevery(10,0,1,2,3,4,5,6,9).weave().trim(0,2935)

return(p3)

That's way more complicated than is usually necessary.

Have a look at http://avisynth.org/mediawiki/ShowFiveVersions

I tried that and it doesn't seem to work for me.

You may just need to force the correct field order.

Try avisource(...).assumebff().separtatefields() and examine the motion. Does it go backwards at all? If so try avisource(...).assumetff().separatefields().

Moth3r said:

You may just need to force the correct field order.

Try avisource(...).assumebff().separtatefields() and examine the motion. Does it go backwards at all? If so try avisource(...).assumetff().separatefields().

That did it!  Thanks for the advice.  I'd have never figured that out on my own. 

I'm done with side 1 now.  I'd like to encode a few minutes worth and post it to get some opinions.  Does anybody know where I could post a raw file or should i just upload it to vimeo?

I'm only using the composite tooted average now because I was seeing too much dot crawl in the s-video captures.  It looks like this is going to be raw other than the toot and ivtc.

Here is a 10 minute clip of the beginning of side 1.  There's been no processing other than using toot and ivtc.

Thoughts?

PW = OT