downsampling raster images

massimo
massimo Global Mapper UserTrusted User
edited March 2012 in Bug Report
hello!
I noticed a strange behavior of the different resampling algorithms when downsampling a landsat image.
In particular the nearest and the bilinear ones give exactly the same result, while bicubic gives a result that looks more "noisy" and sharp while I would have expected the opposite effect. I have uploaded the result of the resampling in a zip file (resampling_test.zip).
Regards,
Massimo

Comments

  • global_mapper
    global_mapper Administrator
    edited March 2012
    Massimo,

    When you say downsampling you mean going to a lower resolution than the original, right? The bicubic resampling is really better for resampling near the original resolution or oversampling, I wouldn't recommend using it if you are down-sampling.

    Thanks,

    Mike
    Global Mapper Guru
    gmsupport@bluemarblegeo.com
    http://www.globalmapper.com
  • massimo
    massimo Global Mapper User Trusted User
    edited March 2012
    Yes, I mean going to a lower resolution.
    I can understand that bicubic does not give good results when reducing drastically the resolution, but why nearest and bilinear should be better? and why the results with both of them are exactly the same?
    I would like to understand it properly so I can make the best choice for the task I have to do..
    Thanks and Regards,
    Massimo
  • global_mapper
    global_mapper Administrator
    edited March 2012
    Massimo,

    I'm guessing that your output pixels were at the exact center of original data pixels, so the bilinear and nearest neighbor resampling would be identical since sampling at pixel centers gives the same results for both. It is only if your samples are off-center that bilinear would make a difference.

    Bicubic will take into account the nearest 16 samples and will cause smoother inter-pixel transitions when over-sampling, but won't really be useful when sub-sampling. If you leave the interpolation at the default Global Mapper will automatically choose a good resampling method when sub-sampling. For example if significantly down-sampling your data a box averager of the appropriate size will be chosen so that all source pixels inside a new output pixel will be averaged to create the new one rather than just using those near the center of the new sample location.

    Thanks,

    Mike
    Global Mapper Guru
    gmsupport@bluemarblegeo.com
    http://www.globalmapper.com

Categories

In this Discussion