What are resize filters

[Gladiator:)]

Actually can someone explain me what resize to use where

Where to use lancoz , what is it ?:confused::confused:

Where to use Bicubic ? what is it?:confused::confused:

Where to use spinline ? what is it ?:confused:

ETC

:thanks:

[unskinnyboy]

http://avisynth.org/mediawiki/Resize

[Blue_MiSfit]

Ding!

And, for the lazy - in brief -

A resize filter is a filter to scale / resize / stretch a video to an arbitrary size - using any variety of algorithms, each with their own advantages and disadvantages.

For example, lanczos is considered "sharp", whereas bicubic is considered "less sharp" (and bilinear is "soft"). Spline is usually considered to be "neutral".

The differences are typically quite subtle - even to those of us with good eyes for video!

Experiment with a few resizers. They all have a unique flavor :D

~MiSfit

[TheRyuu]

I prefer Spline for downsizing and upsizing, I've compared it with the supposedly sharper Lanczons resize and I see no difference, even zoomed in at 400%. Supposedly spline has less haloing so I just use it as a general everything resizer unless something needs special treatment.

[Blue_MiSfit]

And guys - it's Lanczos

:)

[fibbingbear]

For a graphics class in college, we had to implement some filters... you can get an idea of what some of them look like at this collage.
Scroll down to reconstruction filters. Not exactly the same as the ones used in avisynth, but should give a good idea:

http://ybit.org/pewpew/examples.html




A reference image can be found at:

http://ybit.org/pewpew/hidino.jpg

Reference image is supersampled and then downsampled using Lanczos --- for example, the horizon should be a straight line, not jaggy. So the reference image has some artifacts from the Lanczos filter, but it does look sharp.

[unskinnyboy]

fibbingbear, was the supersampling intended to anti-alias the reference image? I ask, because the resized images have jaggies too. When you are downsizing by ~2x, was the supersampling needed at all? What kind of supersampling was used? A simple LanczosResize of the reference image gave me this:

http://img178.imageshack.us/img178/5584/clipboard01po4.th.png (http://img178.imageshack.us/my.php?image=clipboard01po4.png)

..with no jaggies.

[fibbingbear]

The assignment was a ray tracer, so we had a description of an image (for the dinosaur, it was a mesh), and had to render it.

For the reference image, each pixel is a ray cast. I think it was super-sampled at 2x, so first an image of 1280 by 960 was made. However, there are some artifacts present because more rays were needed, even at that resolution (technically may need an infinite number...). You can mask these artifacts by down-sampling. It was down-sampled by 2x using a Lanczos filter.

For the group of 6 images, I think they were rendered with 512 by 512 rays. They were then down-sampled using their respective filters. Since they were rendered with much fewer rays, they have more artifacts in them. I had to use 512 by 512 rays when submitting the assignment due to time constraints.

So in other words, those 6 images demonstrating the filters were NOT based on the reference image. I shouldn't have called it that in my original post. They are based on a 512x512 image that is not on that website.

And I still see minor jaggies in your resize :)