Star Wars Laserdisc Preservations. See 1st Post for Updates. — Page 17

About two captures, I mean IF it's possible to switch phase, of course... like the following:

Actually is a real trick, as first and second frame are sequential, but it shows that it's possible in theory...

Here's a sample of the result of low pass filtering THX 1138's chroma to 843.75KHz (that's 1.6875MHz together which is basically DV's 4:1:1).

This isn't what LD is capable of (>1.3MHz for I and Q), but it's double S-VHS and the rainbows disappear, so it's good enough for me!

http://screenshotcomparison.com/comparison/125913

Note: No 3D comb filter was used here. This is 100% spatial.

WOW!!! Great!

Can you explain the method to get this kind of results?

_,,,^..^,,,_ said:

WOW!!! Great!

Can you explain the method to get this kind of results?

It's really simple, actually. A low pass filter is basically a blur.

NTSC rotates the phase such that you have Y-I Y-Q Y+I Y+Q, so if you average (horizontally) across enough samples, you get rainbow cancellation at the expense of detail loss in the chroma, but our eyes aren't very sensitive to chroma, so this should work well enough.

In order to do this, I assumed a sample rate of 13.5MHz over 4:2:2 sampling (true for my setup, at least) which is 6.75MHz for luma and 3.375MHz for each chroma channel. This works out to ~106.67 pixels per MHz which implies that I need to reduce each chroma channel from 360x480 to 90x480 in order to achieve 843.75KHz.

(S-VHS chroma is 400KHz, so this is surprisingly more than double that!)

The Avisynth for the sample I provided is: thx.SeparateFields().Spline16Resize(180, 240).Spline16Resize(720, 240).Weave().MergeLuma(thx)

Do note that I use 180 instead of 90 here because that's the luma width (I put the original luma back in with MegaLuma at the end), so the chroma width is half of that: 90.

Also note that I SeparateFields() and Weave() here even though it may not be necessary because I didn't take the time to examine the Spline kernel and I wanted to be safe.

happycube said:

Nice, that looks very effective!

Thanks!

@Andrea: 843.75KHz is no magic number. I just chose it after some experimentation and enjoy that it's equivalent to DV's 4:1:1 horizontally, at least.

You can, for instance, get 675KHz with 72 (e.g., Spline16Resize(144, 480)) instead of 90.

675KHz gets pretty close to "narrow band" NTSC.

Do note that this is only for NTSC. You'll have to work out other numbers if you want to try it with PAL, but it should be the same procedure.

@AntcuFaalb: thanks for the info! Actually, I prefer to use a derainbow filter that, even if is not that effective, at least leaves the chroma at its native resolution; even if I agree that we are less sensitive to chroma than luma, it's always true that laserdisc chroma is around 180 pixel (when captured at 720 pixel width) at best - probably could be reached using Happycube software and a good SuperNTSC disc; normal discs could be around 120-140 pixel, that will be reduced to 30-35 pixel with your method, too small for my tastes, and it will be really noticeable when upscaled to higher resolution, in particular 1920x1080.

@happycuber: I'm aware that consecutive frames have different phases, but that "trick" was the only one I was able to do to make an example; the picture could be found here: http://s22.postimg.org/vm22a3rn3/picture_5_X.png - hope you could see it following the link.

I wonder if it's possible somehow to reverse the chroma phase, even with some hardware; here in Italy, if the plug is reversed, the phase does not reverse...

_,,,^..^,,,_ said:

that will be reduced to 30-35 pixel with your method

30-35? 843.75KHz is 90. That's 67.5 TVL. Where are you getting 30-35?

You don't just divide 120 by 4. Capture cards oversample chroma at 3.375MHz. That's 360 lines.

Even if there's only 120 lines of "effective" resolution within the 360, a low pass filter would bring 360 closer to 120 before bringing 120 closer to 0 (or 30 as you claim).

You have to start with the actual resolution (oversampled) rather than the effective resolution when doing these calculations.

Imagine what would happen if you started with 120 vertical alternating white and black lines and then stretched (oversampled) it to 360 and then did this.

Laserdisc effective chroma horizontal resolution will be (at best) around 120 lines, or about 160 pixel, when using the best SuperNTSC perfectly mastered disc, while others have less... that's at native sampling rate of 3.375MHz; when divided by four, we get 40 pixel at best... am I wrong?

_,,,^..^,,,_ said:

and it will be really noticeable when upscaled to higher resolution, in particular 1920x1080.

I've watched an LD and S-VHS tape* (400KHz chroma bandwidth! less than half of this method!) of the same movie on my Pioneer Kuro plasma and I can't tell the difference even up close.

(* I made the S-VHS from the LD.)

_,,,^..^,,,_ said:

Laserdisc effective chroma horizontal resolution will be (at best) around 120 lines, or about 160 pixel, when using the best SuperNTSC perfectly mastered disc, while others have less... that's at native sampling rate of 3.375MHz; when divided by four, we get 40 pixel at best... am I wrong?

You're confusing effective resolution with actual resolution.

You don't have 120 lines of chroma resolution in the file you capture. You have 360. THAT is what's getting divded by 4 here.

Saying "120 lines" in this context means that you can take that 360, resize it to 120, and then resize it back up to 360 without it looking any different (excluding noise).

AntcuFaalb said:

I've watched an LD and S-VHS tape* (400KHz chroma bandwidth! less than half of this method!) of the same movie on my Pioneer Kuro plasma and I can't tell the difference even up close.

(* I made the S-VHS from the LD.)

 ...but I would be sad to know that the chroma isn't there anymore... (^^,)

AntcuFaalb said:

_,,,^..^,,,_ said:

Laserdisc effective chroma horizontal resolution will be (at best) around 120 lines, or about 160 pixel, when using the best SuperNTSC perfectly mastered disc, while others have less... that's at native sampling rate of 3.375MHz; when divided by four, we get 40 pixel at best... am I wrong?

You're confusing effective resolution with actual resolution.

You don't have 120 lines of chroma resolution in the file you capture. You have 360. THAT is what's getting divded by 4 here.

360 are the samples I get, not the effective chroma resolution... I mean, even at 360 pixel width, we will never get 180 pairs of yellow and red vertical lines, for example, just because... they are not there... we'll get 60 pairs of yellow and red lines three pixel wide each, more or less...

_,,,^..^,,,_ said:

AntcuFaalb said:

_,,,^..^,,,_ said:

Laserdisc effective chroma horizontal resolution will be (at best) around 120 lines, or about 160 pixel, when using the best SuperNTSC perfectly mastered disc, while others have less... that's at native sampling rate of 3.375MHz; when divided by four, we get 40 pixel at best... am I wrong?

You're confusing effective resolution with actual resolution.

You don't have 120 lines of chroma resolution in the file you capture. You have 360. THAT is what's getting divded by 4 here.

360 are the samples I get, not the effective chroma resolution... I mean, even at 360 pixel width, we will never get 180 pairs of yellow and red vertical lines, for example, just because... they are not there... we'll get 60 pairs of yellow and red lines three pixel wide each, more or less...

Yes, but that 180 is stretched* over the 360. It's not all in one place. It's not all squished to the left, for instance.

It's more like: [real1] [noise2] [real3] [noise4]... [real359] [noise 360]

(* It's not stretched. It's oversampled. The "holes" inbetween are filled with noise.)

120 lines (max) will be sampled (roughly) as 120 samples 3 pixels wide, that's it... if not, we'll get 360 effective chroma resolution even from VHS, that had (when lucky) around 30...

Also, luma sampled at 6.75MHz gives 720 pixel, that are not the effective luma resolution... I'll kill to have a VHS tape with an effective horizontal resolution of 720 pixel (luma) and 360 (chroma)... don't you? (^^,)

_,,,^..^,,,_ said:

120 lines (max) will be sampled (roughly) as 120 samples 3 pixels wide, that's it...

Exactly! So what happens when you blur every 3 pixels? Do you end up with 120 or 40?

_,,,^..^,,,_ said:

Also, luma sampled at 6.75MHz gives 720 pixel

Sure, but are you forgetting that the effective resolution of LD luma is 425TVL=~567 lines?

567 lines spread over 720. Same deal, just the numbers are more comparable.

It's like the stretch marks on my stomach.  The smaller my stomach gets, the less I see stretch marks :).

AntcuFaalb said:

Exactly! So what happens when you blur every 3 pixels? Do you end up with 120 or 40?

Get the point; the fact is, those three pixels are not fixed; I mean, we could have 2+1/2 red, 3+1/3 yellow, 2+2/3 red and so on... right? Then, that it works good or not I think is dependant also from each disc.

_,,,^..^,,,_ said:

AntcuFaalb said:

Exactly! So what happens when you blur every 3 pixels? Do you end up with 120 or 40?

Get the point; the fact is, those three pixels are not fixed; I mean, we could have 2+1/2 red, 3+1/3 yellow, 2+2/3 red and so on... right? Then, that it works good or not I think is dependant also from each disc.

Yep, it definitely depends on how the disc was encoded. It can't kill certain rainbows in the JSC, for instance. The chroma bandwidth is really, really wide on that one and it dips into the lower frequencies.

Made a brief test, and it seems to work quite well... despite the fact my brain is still convincing me that the contrary is true... but it need further investigations, nevertheless! (^^,)

Here you are your avisynth function:

function lowchromanoise (clip clip) {
clip
SeparateFields
Spline16Resize(clip.width/4,clip.height/2)
Spline16Resize(clip.width,clip.height/2)
Weave
MergeLuma(clip)
}

don't know if using other spline, or different resizer at all, will increase the quality, though... hope to find out some other capture to test it...