Lanczos vs Bicubic
- TEKnician
- Joined: Thu Mar 04, 2010 4:40 pm
- Status: Not reading the spoilers or manga
- Location: Smack in the middle of the Pacific Ocean
Lanczos vs Bicubic
Wondering which resizer would be better for upscaling. Other resizers welcome.
From the avisynth wiki, I'm getting the impression that Bicubic is better, but Lanczos has better sharpening? Any thoughts?
From the avisynth wiki, I'm getting the impression that Bicubic is better, but Lanczos has better sharpening? Any thoughts?
Almost as hard as fighting a Holy Paladin.
- Zarxrax
- Joined: Sun Apr 01, 2001 6:37 pm
- Location: North Cackalacky
- Contact:
Re: Lanczos vs Bicubic
Spline36 is generally considered better than either. I think it has less ringing, but it still pretty sharp.
- TEKnician
- Joined: Thu Mar 04, 2010 4:40 pm
- Status: Not reading the spoilers or manga
- Location: Smack in the middle of the Pacific Ocean
-
- Joined: Tue Dec 25, 2007 8:26 am
- Status: better than you
- Contact:
Re: Lanczos vs Bicubic
They're all terrible. The general recommendation is gamma-aware nnedi scaling with spline36 centre correction, run separately for each plane of the video and then recombined. See the Dither 16bit package, nnedi3, and merge() functionality of Avisynth.
- TEKnician
- Joined: Thu Mar 04, 2010 4:40 pm
- Status: Not reading the spoilers or manga
- Location: Smack in the middle of the Pacific Ocean
Lanczos vs Bicubic
THERE YOU ARE. I was wondering when you were going to show up!Mister Hatt wrote:They're all terrible. The general recommendation is gamma-aware nnedi scaling with spline36 centre correction, run separately for each plane of the video and then recombined. See the Dither 16bit package, nnedi3, and merge() functionality of Avisynth.
K, I'm on a Mac using Crossover to run Avisynth and friends.
These packages are obviously separate. So I'll figure out the installation issues. But you say to run for each plane of the video, as in separating the one into several videos, applying a script to each, then merging?
Almost as hard as fighting a Holy Paladin.
- mirkosp
- The Absolute Mudman
- Joined: Mon Apr 24, 2006 6:24 am
- Status: (」・ワ・)」(⊃・ワ・)⊃
- Location: Gallarate (VA), Italy
- Contact:
Re: Lanczos vs Bicubic
No, he's saying to make different clips for each plane within the same script and then merging them. I personally think that it's not too necessary to do, however, and also, nnedi3 is not gamma aware by default, so you will have to either manually correct the gamma before or after the upscale or, just to make stuff easier, use resamplehq with spline36 for luma and sinc with 8 lobes for chroma, which will give a still decently sharp upscale and it will be easier and gamma aware out of the box. Some sharpening post the resize could help but for filters and parameters it's a matter of taste; don't overdo it, but YMMV on what is too much and what isn't.
- TEKnician
- Joined: Thu Mar 04, 2010 4:40 pm
- Status: Not reading the spoilers or manga
- Location: Smack in the middle of the Pacific Ocean
Lanczos vs Bicubic
So most of these suggestions revolve around variations of spline36.
I'll run some tests and do some more research. At least I know which direction I'm going.
Thanks guys
I'll run some tests and do some more research. At least I know which direction I'm going.
Thanks guys
Almost as hard as fighting a Holy Paladin.
- mirkosp
- The Absolute Mudman
- Joined: Mon Apr 24, 2006 6:24 am
- Status: (」・ワ・)」(⊃・ワ・)⊃
- Location: Gallarate (VA), Italy
- Contact:
Re: Lanczos vs Bicubic
Assuming you're trying to upscale an NTSC DVD to 720p, here's basically what you shoud do:
Download link of ResampleHQ if you don't have it.
Code: Select all
mpeg2source("your.d2v")
#ivtc/deinterlace as needed here
resamplehq(1280,720,srcmatrix="TV.601",dstmatrix="TV.709",src_left=8,src_width=-8,chroma_kernel="SINC",chroma_karg1=8)
#sharpening here
-
- Joined: Tue Dec 25, 2007 8:26 am
- Status: better than you
- Contact:
Re: Lanczos vs Bicubic
I actually meant to use dither to unpack the gamma as ResampleHQ breaks chroma at the moment.
The nnedi planar scaling stuff is only really relevant for RGB upscaling however. If you're doing it on YUV, then you can split it into Y and UV if you want but IMO it isn't strictly necessary. The main reason to do channels individually in RGB is that the channels can have different intensity values and these don't always scale the same way, so you tend to not fuck up luma when doing it that way, IIRC. In YUV, the luma is separate so it isn't really an issue anyway.
So I guess you'd be looking at something more like this:
[MOD258: Redacted the code at author's request because it was incorrect.]
If you want to do split channel handling, it's relatively simple by defining a clip and informing nnedi3_rpow2 on which channels to run, then merge back together. If you do this, mirko's suggestion of sinc interpolation for chroma is a good idea. You may also want to remove the nns argument but I put it in to improve speed for negligible quality changes.
The nnedi planar scaling stuff is only really relevant for RGB upscaling however. If you're doing it on YUV, then you can split it into Y and UV if you want but IMO it isn't strictly necessary. The main reason to do channels individually in RGB is that the channels can have different intensity values and these don't always scale the same way, so you tend to not fuck up luma when doing it that way, IIRC. In YUV, the luma is separate so it isn't really an issue anyway.
So I guess you'd be looking at something more like this:
[MOD258: Redacted the code at author's request because it was incorrect.]
If you want to do split channel handling, it's relatively simple by defining a clip and informing nnedi3_rpow2 on which channels to run, then merge back together. If you do this, mirko's suggestion of sinc interpolation for chroma is a good idea. You may also want to remove the nns argument but I put it in to improve speed for negligible quality changes.
- mirkosp
- The Absolute Mudman
- Joined: Mon Apr 24, 2006 6:24 am
- Status: (」・ワ・)」(⊃・ワ・)⊃
- Location: Gallarate (VA), Italy
- Contact:
Re: Lanczos vs Bicubic
What? Also, I hope you're aware that your script can only fix the gamma for the luma and thus YOU have the wrong chroma right there.Mister Hatt wrote:I actually meant to use dither to unpack the gamma as ResampleHQ breaks chroma at the moment.