5

Using Georeferencer in QGIS (which in fact uses GDAL library), in transformation settings you can define the Resampling method as can be seen on the screenshot. I am wondering on what it depends which resampling to choose. E.g. there is best practice for using "Bilinear" Interpolation for elevation/hillshade.

However, when using Georeference, what are the most important criteria to define which resampling method to use? Is it Transformation type? The kind of input image? Other criteria or a combination of both?

enter image description here

Babel
  • 71,072
  • 14
  • 78
  • 208

1 Answers1

4

Every resampling method has different characteristics and is used for different purposes. Nearest neighbour method is used for discrete data because don't assign new values on calculated pixels. Linear is used for continues data as you already said for example elevation. Cubic from my experience returns better results for continues data but its more time-consuming.

In conclusion, there is no universal rule for every situation. It's all about your task, size of raster, computational time and importance of accuracy.

Kostas Pl.
  • 175
  • 1
  • 9
  • 2
    I would add to the list also how much the transformation is changing the image. If the transformation rotates and stretches the image then for example texts in a scanned map will degrade, but not as badly when some smoothing resampling method is used. – user30184 May 24 '23 at 09:09
  • 2
    Have a look at https://gis.stackexchange.com/questions/10931/what-is-lanczos-resampling-useful-for-in-a-spatial-context/14361#14361 for a nice visual comparison between the different resampling options, particularly the graphs. Lanczos resampling can be dangerous, it will tend to push up peaks and dig in gullies if used on elevation data but may give a more visually appealing, but less accurate, result on other data. – Michael Stimson May 25 '23 at 02:59