Making progress on PAL, thanks to samples from Althor1138, CapableMetal, and lddb's megapixie:
(pictures are horizontally stretched since they weren't resized from PAL-4fsc)
http://i.imgur.com/ih5d6Ix.png
edit: (NTSC) 97SE soldier shot earlier posted on another thread: http://i.imgur.com/fUPUZ8f.png
I think you're comparing the spline64's 23755 to your script's output. SRv8's samples are less noisy and it has a shadow error that looks like interlacing.
The GOUT's opening shot is already cleaner than the rest of it, but for PaNup™ it certainly wouldn't hurt!
If the disk was all squeezed, I would be trying to get my hands on at least two copies... PDO UK rot is (a) all to common and (b) completely different than other rotters. On most LD's the signal is destroyed in small spots but the rest of it's just fine usually, but with PDO the reflectiveness itself is reduced all over the disk and SNR is quite low.
So to get a good quality PDO UK rot capture at least two/three copies need to be averaged, whereas a regular disk level 1 or 2 rot usually won't affect my software too badly, the compensation is pretty good. And the merging program when (re)written will reduce rot further if you have multiple copies of the title.
Figured. It would've been so awesome if it was Accidentally Anamorphic, though.
I remember getting an early used copy of On Her Majesty's Secret Service that was, and I have a rather rotted Norma Rae NTSC LD that's squeezed.
It'd be awesome if the whole movie was anamorphic! (But I have doubts, it might cut to regular size after the first shot transition)
AntcuFaalb: Only downside of that is chroma shifts at the same luma won't be picked up. It might be good to search for the rainbows as well I think.
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.
Rather humble compared to what team negativeone can do, but here's a sw97se cap using my software (updated 5/10 - fixed a correction that was overcorrecting, and cropped): http://i.imgur.com/fUPUZ8f.png
And from the JSC: http://i.imgur.com/XYvfeJy.jpg and R2D2: http://i.imgur.com/BR3TjJy.jpg
. o O (re -1's sample: Dang. Even without the color adjustment I look at it and think/whimper "that's what a real SW Blu Ray could look like...")
Yeah - very nice cleanup! (and I added an option to my comb filter to output 844 wide, which should cover everything except the burst isn't getting copied in)
I'm thinking collectively there's a good chance something along the lines of the X0 project could happen again, this time with more stuff/techniques in the open. (How is Laserman anyway?)
Tweaked my software to do color LPF after chroma is added into luma for the actual comb filtering. Didn't get around to doing a wide-output mode, but I figure I'll look at a thing or two more then jump back into PAL.
It sounds like 90 lines to me. The downside I can think of is that the chroma edges can get blurred, resulting in reduced sharpness. But chroma sharpness is not nearly as easy to distinguish as luma - and rainbowing is much easier to spot.
On pre-SuperNTSC disks, there isn't much I information past ~105 lines or Q information above 64. SuperNTSC disks can have up to 2mhz/160 lines.
In theory one could have ~360 lines of chroma but it would become extremely difficult to figure out what luma should be - a 3D filter can do it if there's no movement, but that's about it!
@Andrea: I can't see that picture, but consecutive frames always have different phases, that's the basis of 3D comb filtering.
Nice, that looks very effective! I'll have to look into implementing this in my 2D comb filter, it looks like even with the same # of lines it's a 2D filtering process, and I've only been playing with 1D in that part of it.
(largely for myself later on) Here's some detail about spline16/64/etc - http://forum.doom9.org/showthread.php?t=147117
:)
YIQ decodes will still be RGB48, it's just a different conversion formula. SuperNTSC disks probably won't gain much from YIQ processing, but it might help 70's/80's disks.
I think the TBC is pretty accurate right now - the acid test is the VE zone plate where you have 4-5mhz vertical lines to look at.
It would be possible for the TBC to output 8fsc again, but it would be a lot of comb filter reworking to deal with the in between pixels, and I don't know how much it would actually gain. 4fsc is pretty much the standard for digital comb filtering.
Two LD captures will always have the same chroma phase for each line, it would just reduce noise from the player and ADC. Worthwhile but beneath the noise is the same information.
Thanks (to you and CM :) )
I should be able to add a full-width output option including borders and even the color burst if you want, I'll check up on that tonight. I didn't think that much was missing, but for a real preservation any cropping is bad.
I think the two phase thing is a video processor trick - each (full) frame only actually exists as it's being decoded here. On telecined frames you can average the repeated fields, but without super-precise motion detection which I don't have yet, you would just have occasional frames that look better. Probably better to have that kind of NR done in post-processing.
I'm looking at processing color in the true YIQ color space (wasn't able to make it work last night, but I'll keep plugging away), which may help the JSC since on pre-superNTSC material I is 1.3mhz and Q 0.6, and when working in YUV you can't take full advantage of that. (With SuperNTSC disks you get better results if the LPF is turned off completely.)
Chroma noise is pretty endemic to LD, and the JSC in particular is trying to cram 10lbs of detail into an 8lb bag.
TooT might still be better for absolute signal/noise ratio, but two should be enough to (mostly) de-rot two rotted but otherwise indentical disks. Needs more testing once I have the averager updated to work with the new metadata I'm producing.
As for phase inversion - dunno if that's a real thing or a trick to make the comb filter used go full-3D. I'd have to see an example of it in motion to say.
On my end, I think I'll relatively soon have a complimentary technique to merge two captures of the same master (preferably from different disks) - should be particularly good for rotters (since most of the time the rot will be in different locations) and disks that are pressed a bit noisily (points at Technidisc ;) )
Early PAL TBC test shot: http://i.imgur.com/Dy0qc7T.png
Now to write a comb filter ;) I'm pretty sure I can beat most/all PAL players, for whatever that's worth... (I think the NTSC side currently can reach a hypothetical CLD-97+ with say the 99's imperfect 3D comb filter. No PAL player gets close to that ;) )
Haven't tried yet - I don't know if it will play at the standard PAL speed, and not sure if I can control where it starts playing either, from my experiments the other weekend. But from the manuals it looks like I can run test mode over the serial interface on the V2800, so that might offer a bit more control.
(I also need to clean/replace the grip clamp on my 503-clone before I can say if the HF9G would even try to play a PAL disk in test mode...)
TL;DR I'm doing it as an experiment/lark and not promising anything out of it ;)
The 94 print almost has a grindhouse feel to it - and some of the bits that clip on the THX are better.
Does that THX disk have less EE than the NTSC? Hard to tell on vimeo.
(I need to get cracking on that PAL decoder code!)
It's a later Panasonic design, so it's at least as good as the later Pioneer NTSC midrange players...
... which means it's leaps and bounds better than a 515 playing NTSC disks. Almost all the PAL/NTSC players except the 925 and DVL's (and *maybe* the 2950) have atrociously bad circuitry post-processing NTSC signals.
Others on forum.lddb.com might know more.
Idea (may or may not be easy given the software used) - it'd be nice to have screenshots and descriptions of the disks.
(I don't think there's one LD that "has it all," but if there is I'd definitely like a copy!)
CapableMetal: Sweet :) There's a link on the previous page (and the github readme now) to the google doc with some documentation and a dev log - if I knew you were setting it up I would've made sure you saw it.
sporv: The Core 2 Duo will be fine if slower (my i7-2600 runs at ~10% real time, and optical flow comb filtering bumps that down to 6%), once you adjust the Makefile. I've heard of AC3 mods reusing existing connectors, so it shouldn't be too hard to pull an unused one for RF.
S-Video isn't required on the capture card, just composite. In fact capture quality improved when I powered down the chroma DAC on the board. Tunerless cards probably have less interference and could work better - I actually have one but I use the Hauppauge since it's by far the easiest to get. The card you linked to earlier is the exact card, BTW.
It looks like the phase inversion trick forces the comb filter to go fully 3D. That works very well unless there's any movement. I'm using an OpenCV optical flow algorithm, but the implementation is not perfected yet.
(If either/all of you want edit access to itPM me your google account info. I'm also up for making a thread on fanres :) )
I'm wondering who did the widescreen transfers for the Japanese widescreen version etc, since they came off the same print.
... and why the pre-DVNR version found on THX Wow! and the Technidisc widescreen never got a good quality pressing. IMO that's the best NTSC master...
Very interesting post - and give your dad kudos for his work in the 80's :)