In this example, I've compared six (of eight) different resampling algorithms when decreasing the size of an image.
In this case: a section of a NASA image. You will find the complete image on the NASA Webpage
(3.5 MB).
For resizing the images, the graphic viewer XnView was used.
You will find my personal comments at the bottom of this page.
| Original (section of NASA image): | |
|
|
|
| Nearest Neighbour: | Bilinear: |
|
|
| Lanczos: | Hermite: |
|
|
| Bspline: | Bell: |
|
|
Result: I think, for the used image Lanczos is the best choice.
- Nearest Neighbour: Image looks too sharp, basically.
- Bspline: Makes the image too smooth. Many details are lost.
- Bell: Also very smooth, but not as much as Bspline.
- Hermite: Not as sharp as Bilinear or Lanczos, but better than Bspline and Bell.
- Bilinear: Looks good!
- Lanczos: The best! Compared to Bilinear, it's a little bit sharper.
