DrDre said:
Is rudeness a common feature among former gurus on this forum?
You've gone meta and ad hominem (and, in consequence, nearly circular!) at the same time. I can't figure out which meme to use.
Man... where's TV's Frink when you need him?
DrDre said:
Sharpening vs deblurring = unsharp masking vs deconvolution
Does = in this context mean "is analogous to"?
happycube said:
Also 3mhz is too high for later disks with ~1.5mhz+ chroma resolution.
Somewhere between 1.5 and 2mhz would be good for all disks.
Do you have an NTSC spectrum graph for 1.5MHz I and Q? All I have is the one for 1.5MHz I and 0.5MHz Q.
ray_afraid said:
Any speculation on who got the lightsaber Luke lost at Cloud City and how they got it?
I find it hard to believe that it survived... Either it fell down to the bottom of that shaft and smashed, or went out a vent like Luke and fell to Bespin... and smashed.
Luke's hand became sentient, dropped off the lightsaber to some scary dude, traveled back (or forward?!) in time, and decided to star in a hit TV show.
Current work: I think the best solution to LPFing chroma in an edge-preserving way is to LPF luma to a really low frequency (down below the chroma mix-in point), run a good edge detection algorithm (e.g., Canny) on it, and then not blur across those edges.
We can't use the chroma channels for edge detection because (1) we're LPFing them in the first place and (2) the rainbows will show up as edges.
Thoughts?
A good frequency to LPF luma down to would be ~3MHz.
The reason we want to LPF luma before edge detection is that we want to get rid of the high frequency detail that caused the rainbows in the first place!
NOTE: The LPF'd luma would only be used for edge detection and wouldn't be included in the output, of course...
DrDre said:
I will use EEDI2 on the source, and then use SR to reconstruct the lost detail in the individual frames. Not sure how well it will work, but still worth a try.
That may not work well since SR relies on aliasing. It's effectively a temporal sharpener for high frequency details. When those details fall between the cracks (aliasing) in adjacent frames, they can be recovered using temporal data if the local neighborhood has a sufficiently-high correlation (i.e., little motion). Basically, more aliasing => more details recovered by SR.
EEDI2 is great for AA, but I suppose it'll kill any extra detail gained from SR.
Darth Lucas said:
AntcuFaalb said:
Darth Lucas said:
Lust-In-Phaze said:
I have another small question that I don't know if you've answered before. Do you intend on keeping things like the orange streaks that appear throughout the Tantive IV scene, as well as the cue marks on each reel?
I believe he said he's gonna take those out as the orange streaks are unique to the technicolor prints.
If you're referring to the "tantive orange errors", then this statement isn't true. See: http://fd.noneinc.com/Glitch/starwars-tantiveorangeerrors/starwars-tantiveorangeerrors.html
They're on all Eastman Kodak and IB Technicolor 35mm release prints.
FWIW, many home video releases were made using interpositive prints, so they don't necessarily show up on those.
Ah, you're right. Let me then amend my previous statement to say: "he's taking those out because they aren't on the negative."
Yep, this is true.
SilverWook said:
Someone was bashing that thread on Twitter recently, and used nearly the exact same description. How coincidental!
@DanAshTreeLane?
https://twitter.com/DanAshTreeLane/status/595373660195655683
He seems like a real piece of work as well as a few other things I won't say here. I hope he publicly shames me in response to this. Hell, I might get my 15 minutes for defending SotS! Too cool! I can't wait.
I'm @AntcuFaalb, btw. Read my avatar for further instructions.
DavidMerrick said:
I could get angry at the sheer perversity in this thread but then I remember this forum has a 36-page, nine-year-long topic devoted to restoring Song of the South and this doesn't seem so bad.
I don't see what's wrong with SotS. Yes, it has an inaccurate portrayal of Reconstruction-era formerly-enslaved-but-now-battling-Jim-Crow African-Americans, but it also helps pass on important African-American Uncle Remus folktales and I can't imagine a child forming a negative opinion (or any opinion for that matter) of African-Americans from this at all.
Darth Lucas said:
Lust-In-Phaze said:
I have another small question that I don't know if you've answered before. Do you intend on keeping things like the orange streaks that appear throughout the Tantive IV scene, as well as the cue marks on each reel?
I believe he said he's gonna take those out as the orange streaks are unique to the technicolor prints.
If you're referring to the "tantive orange errors", then this statement isn't true. See: http://fd.noneinc.com/Glitch/starwars-tantiveorangeerrors/starwars-tantiveorangeerrors.html
They're on all Eastman Kodak and IB Technicolor 35mm release prints.
FWIW, many home video releases were made using interpositive prints, so they don't necessarily show up on those.
ilovewaterslides said:
Speaking of cool members, does anybody have news from poita?
I hope he's okay, He hasn't post for a while.
Yep, he's fine, but really really busy. I just spoke with him yesterday.
happycube said:
How is Laserman anyway?
Alive and well.
Rendering 16mm R2A now. It'll take another 4-5 hours.
After that it's on to R2B and then R1 and a few other bits.
How's Usenet and MySpleen?
_,,,^..^,,,_ 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.
_,,,^..^,,,_ 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.
_,,,^..^,,,_ 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:
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.)
_,,,^..^,,,_ 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).
_,,,^..^,,,_ 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:
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.
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.
_,,,^..^,,,_ 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.
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.
Here's the promised sample. I'm just low pass filtering each chroma channel 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!