Image resampling

There is:
Nearest neighbour
Bilinear
Trilinear
Bicubic

And others. But AFAIK bicubic is the best of the above.

John B

You can also use a weighted average of the 4 neighbors by distance.

Another technique is cosine interpolation.

In order of increasing quality (and order of decreasing performance):

point sampling
bilinear
bicubic
n-polynomial filtering
fourrier frequency decomposition
general 2 dimensional DSP filters

quote:

---

Trilinear

---

? On 2 dimensional data sets ?

[edited by - Yann L on April 14, 2002 1:40:13 PM]

quote:

---

Original post by Me
Trilinear
Original post by Yann L
? On 2 dimensional data sets ?

---

Oh - ok. I'm probably wrong. I know where I got it from, but I didn't check anywhere - obviously I should have done.

John B

[edited by - JohnBSmall on April 14, 2002 1:54:41 PM]

Well, just to clarify: trilinear is the same for 3D images than bilinear for 2D ones. Bilinear interpolates between 4 pixels, trilinear between 8, just a dimension more. There is also the tricubic interpolation method, this is just as bicubic, but for 3D images.

Lots of people think that trilinear is a 2D interpolation technique, since 3D cards interpolate eg. 2D mipmaps using trilinear. But in fact, it is a 3D method: the interpolation is done between 2 different mipmaps, that can be seen as 2 slices of a 3D volume texture.

/ Yann

Where can I find the afaik bicubic algorithem?

Waiting for reply.
Thanks in advance.

http://www.google.com/search?hl=en&q=bicubic+interpolation

First match just happens to be "Bicubic interpolation for image scaling"...

By the way (BTW), "AFAIK" isn't part of the algorithm name, it's an internet acronym for "A s F ar A s I K now"...

Some others:
IIRC = If I Remember Correctly
IMO = In My Opinion
IMHO = In My Humble Opinion
YMMV = Your Mileage May Vary
OTOH = On The Other Hand
FWIW = For What Its Worth
FYI = For Your Information


--
Simon O'Connor
Creative Asylum Ltd
www.creative-asylum.com

[edited by - S1CA on April 15, 2002 10:09:44 AM]

to demo a few really good algos, goto http://www.cs.washington.edu/research/imagedatabase/demo/resample/

basically the best image resampling techiques involve using edge detection in an attempt to recontruct edges to try maintain sharpness of the image when interpolating missing data. this means all the detail presented in the original image is not blurred away like when using bilinear interpolation. bicubic is better then bilinear, but also is done across the image ignoring edges.