Posts: 5
Threads: 1
Joined: 2017 Jan
Thanks: 2
Given 0 thank(s) in 0 post(s)
Country:
2017-01-01, 12:24 PM
(This post was last modified: 2017-01-01, 12:34 PM by Darkbluesky.)
I would like to try this with a Laserdisc cartoon, trying it in avisynth for example, but although I have read all the topic twice, I have still one doubt.
I can capture it to 720x576 and 720x480 (is it better 640x480?), but if then I use Average(), both clips should be in the same size, moreover, if the final frame should be 1080p how can I get that (576+480=1056, and 1920 of horizontal size?). Is it so rough/simple as to resize (with, for example with SuperResolution) each source to 1080p, and then make Average() with 0.5 weight for both? Well, I am pretty lost on that: how should I manage the sizing of the sources and the avisynth Project size? Please, could you explain me it a Little?
Thanks!
Posts: 2,252
Threads: 82
Joined: 2015 Jan
Thanks: 488
Given 845 thank(s) in 440 post(s)
You need to have two independent sources, not the same LD capped twice. Capture it at the maximum width (768 if possible) and 480 for NTSC or 576 for PAL.
Posts: 5
Threads: 1
Joined: 2017 Jan
Thanks: 2
Given 0 thank(s) in 0 post(s)
Country:
2017-01-01, 02:00 PM
(This post was last modified: 2017-01-01, 02:13 PM by Darkbluesky.)
(2017-01-01, 01:57 PM)Valeyard Wrote: You need to have two independent sources, not the same LD capped twice. Capture it at the maximum width (768 if possible) and 480 for NTSC or 576 for PAL.
Thanks Valeyard
Ok, and upscale -the one with best quality for example, or the one with more info on the borders- to 1080p, then upscale the other the factor needed to align it correctly over the first, then interleave (maybe not needed, I have to test), and average () ?
I have both, the NTSC and PAL version of the LD. I guess I'll have to adjust color first in order to have both to be the more similar possible.
Posts: 2,252
Threads: 82
Joined: 2015 Jan
Thanks: 488
Given 845 thank(s) in 440 post(s)
Oh OK, then yes you need to resize the lower quality source first (probably the PAL) to match the other. If the colour doesn't match though they might be separate transfers that won't align.
Posts: 5
Threads: 1
Joined: 2017 Jan
Thanks: 2
Given 0 thank(s) in 0 post(s)
Country:
2017-01-01, 02:18 PM
(This post was last modified: 2017-01-01, 02:20 PM by Darkbluesky.)
Ah! ok, you mean then to resize and align both before upscaling to 1080p, it isn't? Then upscale and average().
Posts: 2,252
Threads: 82
Joined: 2015 Jan
Thanks: 488
Given 845 thank(s) in 440 post(s)
No, you average before upscale. I do have my own script for this (somewhere). You don't use average - you use something like repair or similar. Basically you improve the better 480i or 576i source before up-scaling.
Posts: 7,153
Threads: 601
Joined: 2015 Jan
Thanks: 1081
Given 1466 thank(s) in 963 post(s)
Country:
Nope, you should align AFTER upscale, to have better alignment - one pixel upscaled precision is better than one pixel not upscaled...
Usually, the two versions (PAL and NTSC) has slightly different image size; sometimes one has more on the sides, while the others more on top and/or bottom etc. - an over-exagerated example is this - just to get an idea:
🔍
Now, let's say the green is PAL and red is NTSC; you have to upscale the PAL in the vertical direction to get the "right" aspect ratio - for example, a 2.35:1 movie will be 816px height, a 1.85:1 1040px, a 1.33:1 1080px etc. - for the moment, you will use a temp horizontal resolution equal to the 1080p right width - 1920px for 1.78:1 and higher, 1440px for 1.33:1, and other for the AR in-between.
Then, upscale the NTSC horizontally with the right width, and temp vertical resolution will be the same of the PAL. Now the "fun" part; you have to "shrink" vertically the upscaled NTSC, until they align with the upscaled PAL; but at the same time, you have to "stretch" horizontally (and shift accordingly) the upscaled PAL to align it with the upscaled NTSC... it's a trial and error procedure, but sometimes it worths the time (sometime not...)
WARNING: you will come up with four "dead" corners; if they are small (few pixels) you could try to copy and paste zones from nearby parts; if they are bigger, you can try to use an inpaint script.
At the end, you must find the correct ratio for the average - for example 60% NTSC/40% PAL - to get a perceived better quality of the result in comparison to the single, better version.
Colors: if they are not too far, you can average them, too; at the contrary, you could always use one version chroma, and average luma.
Posts: 5
Threads: 1
Joined: 2017 Jan
Thanks: 2
Given 0 thank(s) in 0 post(s)
Country:
Thanks. I think thsi is clear now for me.
I'll try it and see what I can do. BTW, I understand that if both sources are captured at 720 width, only (mainly) vertical improvement is to be expected. That it means that it is preferable to capture NTSC at 640 width?
Posts: 7,153
Threads: 601
Joined: 2015 Jan
Thanks: 1081
Given 1466 thank(s) in 963 post(s)
Country:
It's always preferable to capture at native card resolution - usually 720 horizontally, and 480 or 486 for NTSC and 576 for PAL vertically.
Even if the image width is the same (e.g. they match horizontally), probably the analog dots are placed in a slightly different way, so it's worth to give it a try - at the end, vertically they ARE different!
Don't forget to post here some screenshot comparisons - best version (PAL or NTSC) Vs. PaNup!
Posts: 2,290
Threads: 39
Joined: 2015 Jan
Thanks: 177
Given 187 thank(s) in 133 post(s)
A couple of the projects I'm working on right now involve upscaled footage taken from two different SD sources (one NTSC and one PAL) and I agree that the best results (as far as getting the images to align properly) are AFTER you have upscaled the two sources, as there is a better precision because you are dealing with a bigger image.
|