Is this enough to warrant 1080p?

[Comatose]

1080p Blu-ray: The difference between the spline36-resized 720p->1080p clip and untouched 1080p is so small that I'm not sure the jump in video resolution is worth the small increase in actual resolution (ignoring the grain).

Make sure your browser isn't scaling the images down.

Source (http://91.121.134.114/compare/orig.png)

spline36resize(1280,720).spline36resize(1920,1080) (http://91.121.134.114/compare/resized.png)

--

source.pointresize(width*2,height*2) (http://91.121.134.114/compare/orig-point.png)

source.spline36resize(1280,720).spline36resize(1920,1080).pointresize(width*2,height*2) (http://91.121.134.114/compare/resized-point.png)

The difference is a lot more obvious in the pointresized version, especially on the pinky nail of the girl closest to the camera.

What would you do? Would you keep this 1080p, or downsize to 720p?

[poisondeathray]

I would look at some other frames and sections. 1 frame wouldn't sway me either way

[Comatose]

They're all about the same as this one. This is actually a relatively extreme case, most of the rest show even less difference.

[IgorC]

Depens on source. This particular anime source hasn't too many true details.

[Keiyakusha]

My choice is 720p. I don't see any additional details in 1080p compared to downscaled image. Maybe lines a little bit softer (with spline36resize(1280,720).spline36resize(1920,1080)) but with sharper resizer it may be the same.

[tph]

The 1080p Blu-Ray is most likely a slight upscale from the original source, which is quite common with anime. I'd say stick with 720p.

[Gokumon]

The 1080p Blu-Ray is most likely a slight upscale from the original source, which is quite common with anime. I'd say stick with 720p.

Yep very few digital anime are rendered at full 1080p.

To answer the question, it's only worth it to keep it at 1080p if the quality of the upscaling of whatever you have playing the file is going to be worse than just leaving it at the original resolution. If you are playing this on any standalone device many of them aren't going to provide quality comparable to spline36resize it will usually be a bilinear or bicubic resize. If you are just playing it on your pc and you control the resize algorithm, then you are probably perfectly fine downsizing it to 720p and resizing it on playback.

[Keiyakusha]

But with additional processing wich IMHO this source really needs like denoising and sharpening, 720p should look not worse then untouched source even with bilinear.

[*.mp4 guy]

dx=1280
dy=720
taps=31

lanczosresize(DX*2, DY*2, 0.5, 0.5, taps=taps)

pointresize(DX, DY)

spline36resize(1920, 1080)

This applies a downsize to 720P without lowpassing the image first (Simplified explanation) and then upsizes back to 1080P. This example image (http://www.imagebam.com/image/ab278938735470) made with this script would indicate that the source is not of greater then 720P resolution (if it was there would be aliasing, also, all source detail apears to be kept perfectly). The number of taps can be lowered for faster, and slightly worse performance (I didn't test what number was actually needed in this case), also, the source apears to have been improperly resampled at some point, as the lines have regularly spaced discontinuity artifacts.

[Comatose]

Awesome, a reply like that is just what I was hoping for. Can you please elaborate on the discontinuity artifacts? I'm not really sure what you mean.

[*.mp4 guy]

It looks like the resampler they used is somehow non-continuous, the only avisynth interpolators that can cause such artifacts are pointresize and gaussresize (with certain values of P, iirc). They aren't all that bad, and fixing them is quite dificult, so don't worry about it if they don't bother you.

[10L23r]

why did you choose taps=31?

edit: shouldn't this thread be in the avisynth usage forum???

[Comatose]

No. I felt it concerned compression more, and since H.264 is my format of choice, I posted here. I don't think it really fits in any of the forums.

Thanks *.mp4 guy.

[*.mp4 guy]

why did you choose taps=31?

Because the source has no high frequency information close to the nyquist frequency (having been interpolated already) high taps should not cause ringing on it, and high taps have better frequency response, which means less blurring.

[Trahald]

Since this thread has mostly (90%+) dealt with avisynth, I agree it belongs there. It may not have been what you intended but its certainly ended that way. The question can apply to any format capable of 1080p

[Keiyakusha]

dx=1280
dy=720
taps=31

lanczosresize(DX*2, DY*2, 0.5, 0.5, taps=taps)

pointresize(DX, DY)

spline36resize(1920, 1080)

This applies a downsize to 720P without lowpassing the image first (Simplified explanation) and then upsizes back to 1080P. This example image (http://www.imagebam.com/image/ab278938735470) made with this script would indicate that the source is not of greater then 720P resolution (if it was there would be aliasing, also, all source detail apears to be kept perfectly). The number of taps can be lowered for faster, and slightly worse performance (I didn't test what number was actually needed in this case), also, the source apears to have been improperly resampled at some point, as the lines have regularly spaced discontinuity artifacts.

Is there a way to increase the amount of aliasing or something? Because in the cases where upscaling is not that huge as 720p->1080p, its hard to tell the difference.