Difference between Resize filter mods?

[junglemike]

Hi everyone, I was playing with Resize filter in Dub. it has
Nearest neighbor,bilinear,bicubic and other mods.
Can anyone explain what is the difference between them.
I heard that come "rise" compressibility of file, but reduce quality(starirs effect) or vice versa. Is there some method that is good for visual quality and for compressibility too? And Is there some external filter that does it better, or just in other way?
Any help is appreciated.

[hakko504]

To begin with you can always read the help file in VD. It is explained quite well there.

[^^-+I4004+-^^]

"Precise bicubic uses a full bicubic filtering kernel rather than a 4x4. It works very similarly to precise bilinear,......."

>To begin with you can always read the help file in VD. It is explained quite well there

yes,now he (same as me) knows a whole helluva more about resizing than before....LOL!

no,those are not good explanations.....
those explanations are from developer to developer....

[Evil Andy]

K, I'm a little pickled so I don't know how coherent this will be...

Let us consider a picture being resized down by an arbitrary amount. Therefore each pixel in the final picture will be equivalent to several pixels or part of pixels from the starting picture.

Nearest neighbour is the fastest algorithm because it chooses the pixel that would be closest to the original, spatially, if you scale up or scale down an image with no reference to the colour of the surrounding pixels e.g. if you have 9 pixels arranged thus

1 2 3
4 5 6
7 8 9

and you scaled the image down to a third of its size then the resultant pixel would be the same colour as pixel 5 regardless of the surrounding pixels.

Bilinear takes all the source pixels that make up the new pixel and gives a weighted average of their colour to find the final colour. It does this in a linear way ie. a pixel twice as far away has half the weight, a pixel four times as far away has a quarter the weight etc. so a graph of proximity versus weight gives a straight or linear line.

Bicubic is the slowest algorithm as pixels have exponentially more weight the closer they are to the nearest neighbour pixel so a graph of proximity versus weight gives a curved line.

And that all seems to make sense after 7 pints. Stand by for an edit in the morning when I realise I'm talking total rubbish :)

[Evil Andy]

To answer your other question, as a general rule of thumb bilinear should for the most part give the best compression as it throws away the most detail and bicubic should give the worst compression as it keeps the most detail Andy said as he slid drunkenly under the table

[fccHandler]

From my experience, "Precise bilinear" compresses the best of any mode, but it's a bit blurry. "Precise bicubic (A=-1.00)" is the sharpest, but it produces more "ringing" artifacts.

[^^-+I4004+-^^]

to be precise,"ringing" is produced by codec,not resizer.......

bicubic is sharper,that's why.....

[Asmodian]

I though bicubic on something as sharp as -1 produced it's own ringing artifacts, which is the reason lanczos resize exists?

[fccHandler]

No, he's right. It's the codec.

IIRC Lanczos was added only because so many people requested it. Avery Lee said that he couldn't see any difference at all. (I agree.)